We recently had the chance to tour Express Manufacturing, Inc. (EMI), an electronics manufacturer based in Southern California. Touring a shop like EMI is a reminder that electronics aren’t “made” so much as they’re assembled, qualified, and proven across dozens of tightly controlled steps. Once you see that system up close, it’s easier to understand why building reliable electronics at scale is hard, and why the shops that do it well are worth studying.

What follows is a factory-floor walkthrough, equal parts education and the day-to-day mechanics of electronics production. 

About Express Manufacturing, Inc. (or just EMI)  

EMI is one of the largest electronics manufacturing services (EMS) shops in Southern California, founded in 1982 and grown over four decades into a six-building network across the region. It’s still family-owned, and the scale shows: this facility runs everything from PCB assembly to full product assembly, along with installation, test, and whatever QA a customer’s program requires. Despite operating only two shifts with a little over 100 employees, they shipped close to a million units in a single month last calendar year — a reminder of how much volume a well-run, high-mix EMS operation can push through a single site.

Where It Starts: An Idea, Consignment, or Turnkey

The flow to work with EMI is simple for those familiar with buying manufacturing services. Companies can choose consignment or turnkey for fabrication, and also make use of engineering value-add services to boost the bandwidth of internal engineering departments.

With turnkey manufacturing, companies allow the shop to excel at what they do best; that is, sourcing parts, managing vendors, and running builds end-to-end. With consignment, customers supply some or all of the materials, and the manufacturer charges for labor, processing, and assembly. In this model, you retain ownership of the supply chain and take on more of the integration risk.

Most often a supplier like EMI can do a better job of buying components, getting boards made, and designing test fixtures than you can. For example, they have a longstanding relationship with Summit Interconnect for board fabrication (see our tour of Summit’s Hollister facility here).

That said, there are cases where consignment is required or the best option available. If customers are using controlled components a manufacturer can’t source, you have to consign them. If a program requires a company to stick with certain subcontracted suppliers—because of internal agreements, regulatory constraints, or customer requirements—consignment is the only path. And while it’s less common, sometimes companies genuinely have a cost advantage and can procure certain parts or services cheaper than the CM. 

Then there’s the case where companies don’t actually have much engineering work done yet and instead bring a concept for the product. In that situation, it’s usually best to lean on as much turnkey support as possible. Let the contract manufacturer do what they’re optimized for, and free up internal engineering bandwidth available for parts of the design that truly need it. 

No matter which path you choose, all paths point to the shop floor in the end. Specifically, the design for manufacturing (DFM) and new product introduction (NPI) departments. These teams make sure the design is ready for production and chart how the product will flow into the equipment and processes the factory actually runs.

Types of Soldering Machines

For most jobs, manufacturing at EMI starts with soldering. PCBs and mechanical enclosures are fabricated off-site by other vendors, so before the PCBA can be mounted into the enclosure, it first must be assembled. That means populating the bare PCB with components and soldering them in place. In practice, it’s exactly what you’d expect: parts are picked, placed, and reflowed onto the board to turn a piece of fiberglass into a functioning electrical system that can then be integrated into the product.

Reflow Oven

Reflow ovens are the most common type of soldering method in high-volume assembly and how most components are assembled to boards. Solder paste is applied to the PCB via a stencil, and then the PCB components are robotically placed onto the board. At each stage, there are inspections: the paste is inspected once it is applied, components are inspected once they are placed, and the end PCBA once it is soldered.

There are a lot of finer points to running reflow ovens like: 

• The component.  Components have to be fed in by reels, and there’s a limit on how many reels (and thus unique components) a machine can be fed. EMI has a great system of loading the components onto a cart, and then if a project needs more than one cart, they have it loaded and ready to be used.
• Component attrition during placement. Electronic components cannot be removed from the reel and placed on the board with 100% yield; some fall off, and are often too small and fragile to be recovered. The manufacturer has to slightly overestimate the component count by ~5% (depending on the component) to have enough for the order.
• Understanding the optimal thermal profile. Especially for applications like aerospace and defense, where a combination of high temperature and vibrations can be expected, engineers will specify the right type of solder that can activate to be strong enough for the job. The solder has to be stored at the right temperature, and most importantly, the oven has to ramp up to temperature at a certain rate and also cool at a certain rate to ensure effective wetting of the solder.
  • Wetting is the term used to describe how solder fuses into the crystal lattice of both the component lead and PCB.

Automated assembly covers the bulk of the board assembly. Double-sided designs require a second pass after flipping, and a small set of parts — typically large, heavy, or heat-sensitive — are installed manually downstream.

Wave Soldering

Wave soldering is how thru-hole components are installed. Once the industry standard, this method of soldering became less popular once surface-mount components were introduced as the industry miniaturized.

With wave soldering, ingots of solder are melted into a bath of molten material. The level of the solder is critical; surface tension is used to keep the liquid solder in contact with the PCB as it moves over the top of the solder path. This allows the solder to completely wet the component lead and wick into the barrel of the thru-hole, without flowing onto the surface of the PCB above.

Selective Soldering

Selective soldering is used when a mix of thru-hole and surface-mount technology is in use. However, selective soldering is only used to install the remaining thru-holes after the PCB goes through the pick-and-place process and reflow oven.

In selective soldering, a smaller bath of solder is maintained and then robotically actuated in a path that solders all of the thru-holes while avoiding the surface mounts.

Hand Soldering

Hand soldering is used in applications where no other method of assembly is suitable. Often it’s just a handful of components that need manual care, like heavy components on a two-sided PCB that might fall off, sensitive components that cannot handle the thermal profile of the solder oven, or custom components that require additional inspection. EMI has what is by far the largest hand soldering capacity we’ve seen from a manufacturer, so there is probably no hand soldering job which is too big for them to handle.

De-Panelization

During the entire assembly process, the pick-and-place machines and reflow oven operate on a larger panel of boards, with boards densely packed for material optimization. The panel is designed to use material as efficiently as possible, and following the assembly process, all of the PCBAs are cut out of the panel.

De-panelization is the most common manufacturability issue EMI sees. If designs don't leave optimal locations for the supports to be cut into the board, then the PCBs may have to be de-panelized ahead of time.

Manufacturers like EMI can accommodate this by designing custom fixtures to hold the boards in place during assembly, but understandably, there’s a lot of efficiency to be gained by offsetting any components that are in a critical position for de-panelization. 

Inspections and QA

Once boards come off the soldering line, the work is far from done. Every assembly step is paired with inspection (some automated, some human, and often both) before a product is allowed to move downstream. At EMI’s facility, entire machines are dedicated to inspection alongside factory operators who perform their own quality checks on both the equipment and the product moving through the line.

AOI: Machine-Run Inspections

One of the biggest differences between a professional EMS and a small shop is the inspection equipment available on hand. Automated optical inspection (AOI) machines perform high-resolution scans of assembled PCBAs, comparing each board against a known-good “golden” reference. AOI can flag issues like missing components, placement errors, polarity mistakes, and solder bridging that are easy to miss by eye.

AOI doesn’t replace human inspection, but it excels at consistency; every board is checked against the same baseline before the product moves forward. 

Flying Probe Test

To validate electrical integrity, EMI uses flying probe testing to check continuity, shorts, and resistance at exposed test points. Movable probes traverse the board without a custom fixture, making this approach well-suited for prototypes and high-mix, low-volume programs.

At EMI, flying probe tests run post-reflow, catching opens, shorts, and resistance issues before boards reach downstream assembly, where failures become more expensive to fix.

Inspection Stations

EMI also operates manual inspection stations where technicians check form, fit, and function using microscopes and bench-top test equipment. These can be simple power-ups to verify basic operation, visual checks to areas flagged by automated test equipment, and randomized quality checks that EMI mandates as part of their zero-defect philosophy. At these stations, technicians can probe solder joints for voids or measure tolerances down to microns.

What stood out during the tour was the integration from AOI to manual inspection stations. Boards flagged by AOI or electrical test route directly to these stations, creating a tight feedback loop that resolves issues without stalling the line.

For many programs, visual and electrical inspection is one of several gates. Products that need to survive real operating environments then move on to environmental qualification.

Environmental Chambers

Environmental chambers verify that the product can meet temperature and humidity specifications. EMI has thermal chambers, some of which have humidity control and can verify that a device functions properly both during operation and outside of operation. These chambers simulate extreme conditions to stress-test electronics for durability and performance. For instance, the chambers can cycle temperatures from -40°C to +150°C while monitoring humidity levels up to 95%, ensuring components and the board itself won't warp, crack, or degrade under real-world stresses.

High temperature testing is especially important for products operating in confined or thermally harsh environments, like automotive dashboards or sealed industrial enclosures, where electronics must survive elevated ambient temperatures. For programs that require it, some of their chambers can combine temperature, humidity, and vibration testing into a single unit, to mimic the effects of mechanical loads. 

Conformal Coating

Once environmental and electrical testing are complete, boards that require additional environmental robustness move to conformal coating. 

Conformal coating is applied after the PCBA has been fully assembled, soldered, cleaned, inspected, and electrically tested. This timing ensures that any defects are identified and corrected before they are permanently sealed beneath the protective layer, while still allowing the coating to be added once the board is confirmed functional and ready for long-term service. In standard flows, the coating step occurs near the end of the electronics manufacturing sequence; typically immediately after final functional testing.

The process involves masking connectors, switches, test points, and other keep-out areas, followed by automated selective spraying, brushing, or dipping to deposit a uniform layer of acrylic, urethane, silicone, or parylene across the area. After controlled curing (UV, thermal, or moisture), a final thickness and visual verification is performed, and the coated assembly is released to the next operation. 

Product-Specific Testing 

Then we get into the fun stuff, with product-specific testing. At this stage, the customer defines the functional requirements, and a test fixture is designed around the product’s interfaces. These fixtures are custom to each program and interface with automated scripts to perform functional verification, boundary scans, firmware programming, and, where required, regulatory checks like FCC emissions.

During the tour, we saw the testing process for a breathalyzer that used different dilutions of ethanol gas to test whether the device can detect the required thresholds of alcohol in someone’s breath, with the fixture automating gas flow and sensor calibration for repeatable results. We also saw in-flight entertainment systems (the screens on the back of airplane seats) set up and tested for burn-in—running 24/7 test images to weed out early failures. And there were numerous product test fixtures used for programming firmware, stress-testing under load, and other functions, all designed collaboratively with the client.

Deep Dive on the Finer Points

How Pick-and-Place Machines Are Loaded

Pick-and-place machines are loaded using carts of components staged ahead of the line. These machines place on the order of 100,000 components per hour and draw from feeder carts stocked with tape-and-reel parts, trays, or tubes. At EMI, carts are wheeled to the machine stations and docked, where vacuum nozzles or grippers align and pick components. Jobs begin with barcode scans for traceability, followed by automatic calibration to account for component size variation, from 01005 passives to larger QFNs. For higher-mix builds, carts can be swapped without fully tearing down the line, limiting downtime between jobs.

X-Ray Counting and Reel Attrition

An X-ray system is used to count components on full reels prior to loading. The system images the reel to determine which tape positions are populated, producing a non-destructive count without opening moisture-barrier packaging. Counts account for typical reel attrition—empty pockets or placement losses—which can range from 2–5% depending on component type. These counts are used to reconcile kitting quantities before builds begin.

How Lines Are Configured Across Facilities

All of EMI’s lines are identically configured across their facilities. That means you can take a process you developed on the lines in Santa Ana and run it directly on some of the overseas lines at their Vietnam and China facilities. This standardization, down to feeder types and software protocols, reduces transfer times from weeks to days, supporting hybrid onshore-offshore models that balance cost and IP security. For global clients, it's a game-changer: prototype in the U.S. for speed, scale in Asia for volume, all while maintaining consistent quality metrics like 99.5% on-time delivery.

Employee Training

EMI does all their own training in-house! The size of the operation means that they train their own IPC instructors in-house and can bring a prospective technician up to speed with the right processes in just a few months. Aligned with IPC standards, these programs blend academic theory with hands-on labs, certifying techs in everything from ESD handling to advanced rework. What sets EMI apart is the customization: modules tailored to client specs, like aerospace vibration soldering, ensuring a workforce that's not just certified but production-ready from the start.

Touring a factory is a reminder that manufacturing, like all engineering, is a discipline of details. The best shops manage the choreography end-to-end, pulling a high-mix design into production while keeping tight control over process, quality, and yield. 

Huge thanks to EMI—and especially Jason and Emily—for opening up the facility and letting us walk the floor!

For technicians and integration engineers on the world’s most complex hardware programs, the gap between “what we designed” and “what actually got built today” is the difference between confidence and chaos.

At Stoke Space that gap became existential. While racing to build a fully reusable launch vehicle with daily design iterations, every off-the-shelf manufacturing software tool failed under real shop-floor pressure. So, the team built their own. That system is now Boltline — and it’s commercially available to any team fighting the same fight.

We sat down with Brent Bradbury, Head of Business at Boltline, to hear how a rocket program inspired the launch of a software business.

“At Stoke, we set out to build a rocket, but even while we were still in early prototyping, off-the-shelf software tools were insufficient to meet our demands. It became clear we needed an engineering toolset that could manage design, engineering, and production of a highly complex, fast-paced project, while offering traceability, repeatability, and audit readiness. None of the tools that already existed met the needs of our complex hardware and fast-paced environment—so we built it ourselves.”
- Brent Bradbury, Head of Business for Boltline

Origins from Stoke Space

How did a rocket program force a software company into existence?

Stoke’s entire bet is full and rapid reusability — both stages flying again with airline-like cadence and minimal refurbishment. Because the goal is to recover and reuse the second stage, it needed an entirely new actively cooled metallic heat shield designed for rapid turnaround after reentry.

“The real bottleneck wasn’t propulsion or materials. It was iteration speed”, explained Bradbury. “When you’re rebuilding, retesting, and reconfiguring hardware daily, any gap between ‘what’s built’ and ‘what’s recorded’ becomes a liability.” Software stopped being an IT concern and became core engineering infrastructure.

Instead of bolting together separate tools for every discipline, Stoke built one unified system from the ground up: the same real-time software that runs both the vehicle and test stands also provides a single data layer tying test data to BOMs and as-built history, with thin, team-specific layers on top. These core elements became Boltline.

What does Boltline actually do to control daily hardware interactions?

At Stoke the cycle is simple and unforgiving: design → build → test → analyze → repeat — often inside 24 hours. At that pace, configuration drift is the enemy. Engineers and techs need instant, certain answers to key questions:

• What was actually built?
• How was it built?
• Which parts were used?
• Where does everything sit right now?
• When was this step completed?
• Why did it behave differently than expected?

“Boltline sits directly in this loop so we can preserve the state and intent across that cycle,” says Bradbury. “It captures the as-built state of hardware while work happens, not after the fact. When a test is run, the data is automatically tied to the exact configuration that was on the stand. Snapshots can be taken before hot fires, reassemblies, or major changes, so results are anchored to reality. Without that, teams burn time reconstructing history before they can even start diagnosing a failure.”

Can you give a concrete example of how that plays out?

One of the clearest examples involves a complex welded barrel assembly that was taking too long to complete. The initial instinct—common in capital-intensive settings—was to add welding stations and expand floor space.

“Instead, we used Boltline to map the work plan in detail and found that the bottleneck wasn’t welding—it was inspection sequencing,” Bradbury recalls. “By moving certain checks earlier, deferring others into downstream test, and reducing redundant sign-offs, we cut the cycle without adding welders, stations, or square footage.”

That kind of decision only becomes obvious when the work itself is visible and structured. Boltline didn’t just record the process; it made the inefficiency legible.

Another example: A batch of washers was found to be out of spec after 10,000 had been purchased. Bradbury explains: “Using Boltline, the team traced where that batch had been issued and identified the specific shelf locations and affected assemblies. Four washers were ultimately found inside an engine. The issue was contained within two hours, the engine was reworked, and a hot fire happened the same day.”

“That kind of speed builds confidence. Confidence keeps people using the system. Usage keeps the data accurate.”

What does using Boltline look like day-to-day for an engineer or technician?

For technicians, Boltline is the authoritative build guide. Work instructions, torque values, configuration steps, redlines from responsible engineers, and sign-offs are available directly where the work happens. QR codes on parts pull up full history back to design, so technicians don’t have to hold details in their head or backfill information later.

Engineers and integration leads keep Boltline open next to CAD. One screen shows what the system is supposed to be. The other shows reality: exactly what’s built, what’s open, what changed, what failed, and the exact configuration at any given time. Anomaly investigations that once required hours, now only takes seconds.

The system runs on tablets and devices already used in the flow of work, instead of forcing people to remember details and backfill them later at a desktop. That matters for adoption.

In theory, any system—even a sprawling network of spreadsheets—can technically track a single washer. In practice, it’s impossible. Technicians simply won’t enter the required information unless the system makes their job dramatically easier. And even if they could, keeping a network of systems connected in a meaningful and scalable way is inoperable.

Broader Landscape

Why do existing ERP and MES tools fall apart on fast hardware?

Traditional ERP and MES systems assume slow cycles, stable products, and heavy overhead. In programs moving at Stoke speed, they add friction. Engineers work around them. Techs keep notes in their heads or on paper. The “system of record” quietly drifts from reality.

“Boltline is built from the shop floor up,” says Bradbury. “It looks and behaves like modern software. Things are where you expect them to be.”

Boltline overlaps with MES but goes far beyond it — upstream into engineering, downstream into test, operations, and maintenance. It replaces disconnected tools with a single source of truth that stays accurate even when hardware changes daily. For advanced hardware programs, it functions more like the digital backbone of the factory.

“Our customers don’t adopt it because leadership mandates it,” Bradbury notes. “They adopt it because it helps them do their job.”

Is this just a rocket problem?

The same dynamics hit aerospace, defense, energy systems, robotics, EV platforms, and biotech hardware. As iteration speeds increase, lagging tools don’t just slow teams down — they create real risk.

Good communication between hardware and software teams isn’t about more meetings. It’s about shared confidence in the underlying data. Boltline links technicians, engineers, and leadership end-to-end so the whole team operates from the same reality instead of negotiating whose version is correct.

What’s the long-term goal?

“It’s simple,” says Bradbury. “Make real-time hardware development the new normal and close the intent-to-reality gap.”

In that future, traceability isn’t something teams scramble to reconstruct after a failure or an audit. It’s ambient—maintained continuously as work happens, so hardware teams can move fast without losing confidence in what they’ve built.

To see Boltline’s shop-first hardware development platform in practice, visit boltline.com/demo.

To cut through the noise, we recently spoke with Kyle Dumont, Co-Founder & CTO of AllSpice, about what his team is building with DRCY and how they’re applying AI to real hardware workflows.

A bit of context: the problem space AllSpice is exploring is design reviews. Traditionally, these are the quarterly checkpoints where teams pressure-test a design against specs and constraints. Design reviews are a non-negotiable part of hardware development because the cost of being wrong compounds with every step toward production. Unlike software, mistakes aren’t easily reversible with a new commit; being wrong means rework, scrapped boards, and the inevitable schedule slip. 

(History has plenty of reminders of what happens when teams aren’t aligned — the Airbus A380 wiring fiasco cost $6.1B in project delays).

With that in mind, it’s useful to look at how modern teams are trying to close the gap between design intent, documentation, and build reality in day-to-day workflows. 

The Current Reality of Design Reviews

Design reviews are still handled the same way they have for the last few decades: schedule a meeting, export a PDF, redline through it on a call, and collect comments from the meeting with information scattered across Slack, email, and shared folders. For those unfamiliar, version control is often just a copied folder on a network drive (hardly sophisticated, but functional). It works for slow cycles and ends up as a single checkpoint before the design moves forward to build or release.

That model worked when the main challenge was simply getting everyone in the same room. The real purpose of a design review isn’t the meeting itself though; it’s making sure the design intent, supporting documentation, and actual implementation all line up. Most issues surface in the gaps between those sources: specs living in one place, datasheets in another, design files exported into PDFs, and decisions scattered across messages and folders.

For all the advances in tools and software, much of hardware development still looks exactly the same to a team from twenty years ago, nowhere more so than in design reviews.

Origins of AllSpice

That gap between modern design complexity and legacy review workflows is the problem space AllSpice set out to tackle. To address it, their approach starts from a fairly grounded premise:  engineers still have to do the hard part with architecture, product definition, and making the tradeoffs that shape a design. That judgment isn’t going away, and no software tool is going to replace it. The opportunity they saw was in everything around it: the tedious, error-prone work of reconciling specs, tracing signals, and digging through hundred-page datasheets to confirm that the design actually matches the intent.

Early on, the first product was a diff engine for ECAD files like schematics, PCB layouts, and related metadata. In simpler terms, it compares two revisions and highlights changes in components, nets, parameters, and layout; not just file differences, but the actual design database.

That solved a basic and persistent problem where most hardware teams couldn’t easily see what had actually changed between versions of a design. By making changes explicit, the diff engine turned revisions into something reviewers could understand incrementally instead of starting from scratch each time.

Once the diff engine could interpret schematic, PCB, and BOM changes at the design-database level, it naturally became the basis for the core platform: central repository revision control for ECAD data. Instead of storing revisions as entire project folders, edits could be represented as structured changes to objects in the design database (components, nets, parameters, geometry, metadata) which can be committed, branched, and merged.

That model made it possible to build reviews directly on top of those commits. A hardware revision becomes a defined change set with a clear before/after state, rather than another exported file. Reviewers inspect the semantic diff, comment on specific objects or regions, and the discussion remains attached to the change history.

In practice, this was a core bet that hardware projects should behave like software repositories, with revisions that are incremental, traceable, and reviewable in context, not static snapshots passed around as folders or PDFs.

Once hardware revisions were represented as structured commits with a reproducible design state, automation became possible at the repository level. Their third product is called AllSpice Actions, essentially a CI system for ECAD repositories —  meaning checks and workflows that run automatically whenever the design changes, like software test pipelines triggered by a code commit.

Typical uses are deterministic checks tied to repository events like running design-rule checks, validating part numbers against approved libraries, verifying BOM properties, generating manufacturing outputs, or flagging rule violations before a review is approved. The results attach to that specific revision, so reviewers see both the human discussion and the automated findings in the same context.

The DRCY Review Layer

What we’ve described so far compresses roughly five years of development into a few paragraphs. In practice, this meant building the platform piece by piece, almost like adding spokes to a wheel around a common data model that could parse schematic, PCB, and BOM data consistently.

Diff made changes visible. Actions made checks automatic and deterministic. DRCY is the next layer on top of that stack: an AI design review agent operating on the same structured design data.

The name is a play on traditional DRC (Design Rule Check), but the distinction matters. Traditional DRC runs fixed tests against predefined constraints. DRCY is meant to reason more broadly about the design during review, using the actual circuit data rather than exported PDFs or screenshots.

Because it works on native ECAD objects, it can trace signals, interpret part relationships, and cross-reference requirements directly from the design database. That avoids the information loss that normally happens when designs are flattened into images or documents for external analysis, where an agent has to infer intent from pixels instead of working from the circuit structure itself.

Under the hood, the system follows an agent-style approach. Instead of running a fixed checklist, it analyzes the circuit and decides what areas warrant closer inspection, using the schematic data, component information, and supporting documentation available in the repository.

For example, if a section of the design involves a power regulator, the review might include pulling relevant datasheet constraints, checking that surrounding components match recommended configurations, and flagging inconsistencies for the reviewer to examine. The goal isn’t to replace deterministic rule checks, but to provide a first-pass investigation that surfaces issues or questions a human reviewer would otherwise have to track down manually.

DRCY is one part of the broader automation stack inside AllSpice, alongside automated checks from Actions and the design history stored in the repository. Because it runs on the same system of record, its findings stay attached to the actual design context, so reviews remain traceable and grounded in the source data rather than separate reports.

In practice, that means flagging incorrect pin connections (required power or decoupling pins left unconnected or not tied together as mandated), out-of-spec power supply connections like under-voltage on required rails, and unconventional but datasheet-allowed pin reuse (for example, special-function pins repurposed as GPIO).

It can also confirm proper handling of NC/DNU pins, power rails, clocks, buses, and interfaces, surface incompatible component configurations or implementations, and catch higher-level logic errors — the sort of issues that usually only show up after someone manually traces the schematic.

Moving Toward Iterative Design Reviews 

What all of this adds up to is a shift in how often designs can realistically be reviewed. On a traditional program, a design review might happen a few times per quarter, usually when the team feels the design is “ready.” Changes accumulate in the meantime, and the review becomes a large, high-stakes checkpoint where many issues surface at once.

What’s been interesting to learn about customers using their workflow is how quickly that pattern starts to change. Instead of long stretches between major revisions, you might see dozens of commits over a short window — sometimes 60 commits across a phase, with 10 or more design updates landing in a single day. Each change carries its validation results and discussion, so review happens incrementally as the design evolves rather than in one large batch at the end. 

Over time, that compresses the review cycle itself. A process that might previously have meant preparing for a major review every couple of months starts to shift toward smaller, more continuous checkpoints. The formal review doesn’t disappear, but it increasingly reflects a history of incremental decisions instead of being the first time the design is examined closely.

Every hardware team wants the same outcome: ship faster, with confidence. Iterative design review is the mechanism that enables it.

Special thanks to Kyle and the team at AllSpice for the behind-the-scenes look at DRCY. See DRCY in action: https://www.allspice.io/drcy-demo

We recently had the chance to see for ourselves, touring one of Summit Interconnect’s quick-turn PCB shops where commercial prototypes are built in just 1 to 3 days. These are the boards used in prototyping and low-to-medium volume production: small batches, tight timelines, and no room for delay. There’s a good chance your prototype boards were built in a shop like this, in the kind of place that cranks out quick-turn PCBs for everyone from weekend tinkerers to massive organizations spending millions a month in R&D.

It’s worth noting that this is one of Summit’s prototype-focused sites, not the high-volume production lines they run elsewhere. What follows is a photo-heavy walkthrough of how a bare PCB gets made from raw laminate to final inspection. If you've ever wondered what happens between “upload Gerbers” and “board arrives in the mail,” this is the process.

Printed circuit boards are fabricated here in Hollister, California, across two facilities totaling nearly 30,000 square feet tucked past patchworks of fields and low-slung hills on the edge of town.

The fabrication process starts with defining the board stack-up: a layer-by-layer specification that determines the board’s materials, thicknesses, and sequence. It’s essentially a recipe for the PCB, and pulled directly from the material traveler that follows each order through the shop.

Think of a PCB as a multi-layer sandwich of copper and insulating material. The copper forms signal traces and planes, while the insulating layers—usually made from fiberglass-reinforced epoxy or commonly referred to as FR4—set the spacing and give the board its structure. These layers come in two forms: core, which is a rigid sheet of fully cured FR4 with copper foil on both sides (like a thin two-layer PCB), and prepreg, which is the same material in a partially cured state that melts and bonds the layers during lamination. Depending on the design, a board might have 4, 8, or 20+ layers.

Walk a few rooms past the raw material racks and you’ll find yourself in a yellow-lit room. It looks a bit out of place, but there’s a very good reason for it: this is the photolithography area. This step defines where copper will stay and where it will be chemically removed. It’s done using dry-film photoresist and laser direct imaging (LDI), a photolithography process scaled for 18" × 24" copper panels instead of wafers.

Dry-film photoresist is UV-sensitive, so the entire room is lit with filtered yellow light to avoid unintentionally exposing the panels. It’s the same reason photoresist rooms in semiconductor fabs have amber lighting.

Inside, you’ll see:

A roll laminator, where dry-film resist is applied to each inner layer copper-clad panel under heat and pressure. This step preps the inner layers for imaging.

A laser direct imaging system where a UV laser scans across the panel to expose the trace pattern. Nothing visibly changes at this stage, the resist simply hardens in the exposed regions.

And after imaging, a wet process line for image development. Here, a mild alkaline solution removes the unexposed photoresist. These are the areas that weren’t imaged by the laser and remain uncured. The exposed areas, hardened by the laser, stay intact and protect the copper traces. Everything else, effectively the unprotected copper, gets etched away in the next step.

Outside the room, you’ll find a massive section of the facility dedicated to wet chemical processes with long rows of conveyorized tanks. Here, the panels enter the etching line where a cupric chloride solution removes the unprotected copper.

After imaging, each layer goes through automated optical inspection where machines look closely for defects like opens, shorts, nicks, pinholes, or leftover debris. AOI systems use high-resolution cameras and structured lighting to scan the copper surface, then compare those images against the original Gerber data. Think of it as a 2D image-based verification process, tuned for fine-line resolution down to 2–3 mils depending on the system.

Any suspect regions (missing traces, extra copper slivers, incorrectly etched features) are logged and presented to an operator for manual review before the board moves on. This is a critical checkpoint before lamination since defects here get locked inside the board and can’t be fixed later.

Before inner layers can be laminated together, the copper surfaces go through a brown oxide treatment step. This process prepares the copper to bond with the prepreg resin during lamination.

Panels are run through a chemical bath that microscopically roughens the copper and deposits a thin layer of cupric oxide (or an equivalent compound). The modified surface has increased roughness and better chemical compatibility, which improves adhesion to the epoxy.

The result is a matte, reddish-brown finish on the copper strictly for adhesion purposes. Skipping this step risks delamination, especially under thermal cycling or mechanical stress.

Layup and lamination take place in a temperature and humidity-controlled area. Prepreg, the fiberglass sheet pre-impregnated with epoxy resin, is sensitive to both heat and moisture. Temperature control prevents the prepreg resin from curing too early, while humidity control keeps the material from absorbing moisture.

To layup a board, operators build the stack layer by layer with alternating inner layers and prepreg, then cap the stack with copper foil for the outer-layer. In prototype and quickturn facilities like this, layers are either manually aligned using alignment pins or are aligned using a pinless lamination system.

The stacks then go into a lamination press, where heat and pressure cure the resin and bond the layers together. If multiple boards share the same stack-up (e.g. prepreg material), they’re pressed in the same cycle.

After layup and lamination, the panel edges are trimmed to a standard size. The result is still just a bare laminate with an unpatterned outer copper layer and no drilled or plated vias.

The next step is to create vertical interconnects (commonly referred to as vias) that electrically link copper features between layers. These are precision-drilled holes that will later be plated with copper to form conductive paths. Without them, the layers are just isolated sheets of copper and dielectric.

A via is just a tiny copper-lined hole that lets signals (or power/ground) travel vertically from one copper layer to another. Some go all the way through the board (through-hole vias), others stop halfway (blind/buried vias), but they all start the same way: either a CNC drill bit cutting precise holes in the stack or, for microvias, a laser ablating them.

Drill size and accuracy matter. Plated holes that carry signals, power, or ground between layers have a tolerance of ±3 mils. Non-plated holes, like mounting points or tooling holes, are even tighter at ±1 mil.

You’ll usually see racks of spinning drill heads or a bit carousel swapping out tools mid-job. Each bit is dialed in for a specific hole diameter, and there might be dozens of sizes per panel.

For even smaller features like in high-density interconnect boards used in smartphones and aerospace, Summit’s Hollister facility uses lasers to create microvias. These are ultra-fine holes, typically just 3–6 mils in diameter and far smaller than what a mechanical drill can achieve. They’re used to connect adjacent layers in dense, space-constrained designs.

After drilling, the panels need to be cleaned and conditioned before any copper can be plated into the holes. The first step is deburring, which removes any leftover fiberglass or resin slivers around the edges of each hole.

Then the board goes through desmear, a chemical or plasma process that removes the thin film of melted epoxy left behind on the hole walls when the vias are drilled. This smear can block electrical contact to the inner-layer copper, so it has to be stripped off to ensure a reliable connection when the vias are plated.

After cleaning, the holes are still non-conductive. That’s where direct metallization comes in. A thin seed layer of conductive material, often a carbon-based or palladium catalyst, is deposited on the hole walls. This creates a continuous path for current so that the electroplating process can later deposit copper inside the vias.

With the holes drilled and a thin seed layer deposited, the board is ready to build up its final copper.

First, a dry-film photoresist is applied to both outer layers and imaged using the same LDI system from earlier. This defines where traces and pads will go, but unlike the inner layers, the board isn’t etched right away. Instead, the exposed copper is electroplated, adding more copper to the trace areas and inside the vias. Immediately after, a thin layer of tin is plated on top. This tin serves as an etch mask.

Standard green soldermask is run on a Seria SSA-DSV650, a vertical double-sided screen printer used for automation. Panels are clamped upright and coated on both sides in a single pass. The machine controls squeegee pressure and speed down to fine increments, which helps maintain a consistent film thickness.

Custom colors get routed to a manual screen printing table. A technician floods the mesh and pulls a hand squeegee across each side. Slower, but flexible, and you don’t have to flush the entire automated line just to change colors.

Legend comes last. It’s traditionally called silkscreen, and it’s applied using a Sprint 300, essentially an inkjet printer for circuit boards. This layer adds part numbers, logos, and labels for test points. As it prints the white ink is tack-cured instantly, so the panel comes out dry and ready for a final bake.

Once the legend is printed and cured, each panel goes through electrical test. This shop uses a flying probe tester, a machine that moves high-speed test needles across the panel to check for opens, shorts, and misroutes. The probes follow the board's netlist and position data directly, which makes it slower than a bed-of-nails setup but much more flexible for high-mix or quick-turn jobs.

Finally, after testing the boards go to routing, where each individual board is cut out of the larger panel, followed by a final inspection before packaging and shipping.

PCB fabrication is a finely tuned process with dozens of carefully orchestrated sequences. We didn't touch on every step, like applying surface finish (ENIG) where exposed copper pads are plated with nickel and gold, or the numerous quality checks embedded throughout the process. But from raw laminate to finished bare board, this is the full process.

Huge thanks to Allison Herrera (our guide and Summit sales rep), Jon Lass (GM of the Hollister facility), and Lisa Holmes (marketing) for making this tour happen and letting us share their story. After walking the floor and meeting the team, including the original shop owner and technicians with over 20 years of experience, it's clear this place embodies the depth of experience that defines so many great North American manufacturing floors.

Academic & Early Career

Growing up in Pakistan, my fascination with technology began with my father's computer, which I saw as an endless source of play and creation. After moving to Vancouver, I acquired a used laptop and taught myself programming. During high school, I developed software tools for personal use and shared them online, occasionally finding unexpected popularity. My first computer was a self-built project, assembled over a year, starting with a motherboard purchased from Fry's Electronics in Palo Alto.

Although I have strong coding skills, I chose to pursue a degree in Mechatronics Engineering at Waterloo instead of a software-focused program. My reasoning was that I could further develop my software development abilities on my own, making a university education more valuable in a field where I had less prior knowledge.

I aimed to expand my understanding in areas that genuinely interested me, particularly physics and mathematics. Initially, I considered studying physics at Caltech because I enjoyed the conceptual thinking it involved. However, I ultimately decided that electrical and mechanical engineering would provide a more practical application of those interests. Mechatronics presented the ideal combination of my existing software skills and my desire to better understand the physical world.

Waterloo's co-op program proved transformative, enabling six distinct internships that shaped my technical understanding across multiple domains. At Blackberry, I developed an internal reporting tool that streamlined engineering project management processes. At Toyota Manufacturing, I created a vision system that tracked tagged objects in 3D space on assembly lines, a patented innovation that monitored tool movements along predefined paths to verify manufacturing quality. My work at Thalmic Labs (which became North, later acquired by Google) provided early-stage startup experience as engineer number four after their Series A funding.

The program also facilitated my research internship at Stanford, followed by two positions at Apple. The first involved detailed mechanical and electromechanical design in a product engineering team, providing rigorous exposure to professional engineering processes. My final internship came after being recruited to Apple's then-nascent autonomous vehicle project when the team was just about 10 people. After interviewing, I returned to complete my academic semester before starting with them. When I returned to begin the internship several months later, the project had rapidly expanded to 50–60 people. This sequence of experiences provided practical exposure across automotive systems, consumer electronics, manufacturing processes, machine vision, and building autonomous systems.

Building a Company

I’ve always wanted to work on something that made a larger impact on the world. Several of my family members work in the medical field, so I always thought I’d start something in that industry, mostly because the tooling is really dated and the problems are evergreen and multivariate. I actually started a company right after college, Scintilla, where we built a contactless ultrasound system using photo-acoustics and vibrometry. I learned that while it’s important to develop at the edge of research and science, it’s just as important to solve a critical need—and to iterate in the market.

Julie and I started working together on an unrelated project during the early days of quarantine in the pandemic. We had a revenue-generating business in six weeks, and I realized that sales and marketing are just as first-principled as engineering. Julie had spent her career selling to large organizations and understanding organizational psychology, so naturally we were a great fit as a team. We ideated around what we'd want to tackle over a longer timeframe and found that building tools for hardware engineers created a perfect overlap of our experiences. My background working directly on these engineering teams, combined with Julie's expertise in enterprise sales, created a complementary foundation for addressing the challenges these technical professionals face.

Dystr

We are building an AI-native computation environment for engineers working on complex physical systems to write, reproduce, and share analysis. Previously, extracting computational value required significant software experience. Now that AI has lowered this knowledge barrier, the opportunity lies in writing, versioning, and managing calculations while keeping them associated with specific projects.

Mechanical, electrical, and manufacturing engineers need to maintain calculations, datasets, and contextual documents within cohesive project frameworks. These teams develop computational and data processing workflows rather than full software applications. Their goal isn't building complex interfaces but creating and using deterministic computations with minimal programming knowledge—such as physics-based calculations or analyzing oscilloscope datasets within discrete projects.

For many customers, learning sufficient code for computational analysis has presented a persistent barrier. They typically resort to MATLAB only in extreme cases, otherwise defaulting to paper, pen, and TI-83 calculators. With Dystr, they perform engineering calculations while maintaining documentation, sharing with teams, and facilitating reviews. This capability extends naturally to automating manufacturing data analysis or any recurring engineering process. Our customers create computational tasks that process periodic datasets, such as measurements from vendors monitoring assembly lines.

The hardware engineering context presents unique requirements compared to software development. Consider field failure diagnostics: Dystr allows engineers to rapidly populate a model with relevant project materials for time-sensitive troubleshooting. This efficiency provides substantial value to customers working under tight deadlines or managing multiple simultaneous projects.

In essence, we're developing an “agentic notebook” or “engineering CRM” where professionals maintain calculations, notes, datasets, and external references while collaborating with AI models on their engineering analysis. This integrated environment transforms how engineers interact with computational tools and project knowledge.

Landscape of Engineering Analysis Tools

The analysis tools available to engineers working on physical solutions haven't evolved significantly in decades. While software development has seen continuous innovation in workflows and tools, engineers designing circuit boards, mechanical systems, or manufacturing processes still rely on outdated approaches. It's surprisingly common to find engineers responsible for critical systems—from aircraft components to automotive safety features—performing calculations on TI-83 calculators and paper notebooks.

At Boeing, for example, engineers routinely spend hours manually processing CSV files from experiments, comparing results through intensive manual effort rather than through integrated computational workflows.

MATLAB was the last significant platform focused on non-programming engineers, but it still requires substantial coding knowledge. Specialized tools like ANSYS or COMSOL address narrow applications but don't solve the broader computational needs.

A critical bottleneck is that existing platforms either demand programming expertise that most engineers lack or offer limited analytical capabilities. This forces organizations into inefficient workarounds: having software engineers build custom tools, maintaining outdated systems, or simply accepting the productivity limitations. The result is significant wasted engineering talent, slower innovation cycles, and knowledge that remains trapped in isolated notebooks or individual workstations.

Current Use Cases & Customer Outcomes

We’ve been surprised by the diversity of engineering fields that are adopting Dystr beyond our initial focus on mechanical and electrical engineering.

Civil engineers working on structural analysis have found significant value in our platform despite using different data types and calculations. One civil engineering firm that conducts forensic investigations has reshaped its reporting process using Dystr. Previously, writing their detailed 100–200-page reports required 40+ hours of engineering time. They've now implemented a workflow that automatically retrieves relevant analyses from their document library based on investigation findings, cutting report creation time by orders of magnitude.

Another unexpected application comes from manufacturers analyzing test data. They've created workflows where measurements from manufacturing partners arrive via email and are automatically processed through Dystr, generating analysis reports without manual intervention.

Major Challenges

Our primary product challenge involves creating interfaces that balance computational power with accessibility for engineers without programming backgrounds. This requires precise design decisions about how to present complex functionality without overwhelming users or sacrificing analytical capabilities.

The investment landscape presents an ongoing structural challenge for companies building software for hardware engineers. Despite recent improvements, capital allocation in this sector remains asymmetric compared to pure software ventures. We've been fortunate to partner with investors like Long Journey Ventures, Omni, and angels who understand the sector's nuances, but this friction persists in the broader funding ecosystem.

Simultaneously, we're navigating the rapid advancement of AI technologies. Each quarter introduces new capabilities that could enhance our product's core functions, requiring systematic evaluation of which advances meaningfully transform engineering workflows versus those offering marginal improvements. This technical evolution demands disciplined development planning to maintain coherent architecture while integrating transformative capabilities.

Long-Term Roadmap

We envision Dystr becoming the central computational backbone for engineering teams working on physical systems. Our platform will evolve into an "engineering team in a box" with a deeply thought-through series of applications designed specifically for technical professionals.

The Dystr environment will progressively handle more routine analytical tasks, freeing engineers to tackle more challenging problems and expand their technical scope. This will create a new generation of engineers who can work across traditional domain boundaries because computational limitations and knowledge silos no longer constrain them.

An electrical engineer might transition smoothly from circuit board design to creating test fixtures for the board to overseeing the manufacturing process, with Dystr providing continuous computational support throughout. This interconnected workflow represents a shift from isolated specializations to more integrated engineering practices.

Automating repetitive work that currently consumes valuable time allows engineers to focus on higher-complexity work and creative problem-solving. The result will be broader technical capabilities for individuals and more cohesive engineering solutions for complex physical systems.

Q&A

Favorite Invention

Instead of focusing on a single achievement, I'm intrigued by the cumulative engineering advances that we often take for granted. For example, modern water and electrical infrastructure have transformed what were once extraordinary capabilities into everyday necessities.

The development of reliable infrastructure is not the result of a single breakthrough; rather, it is the culmination of thousands of incremental innovations in materials science, manufacturing processes, and design methodologies.

What is truly remarkable is not just the technical achievements themselves, but how they have enabled capabilities that would have seemed extraordinary just a generation ago to become commonplace tools. Everything surrounding us—from the devices on our desks to the systems powering our cities—represents impressive engineering accomplishments that we now consider ordinary. This pattern of transforming extraordinary abilities into everyday utilities exemplifies the most impactful nature of engineering.

Favorite Interview Question

I don't have a standard question, but I focus on understanding a candidate's first-principles thinking. When working with emerging technologies like AI, experience with specific implementations matters less than fundamental understanding. I often introduce a basic concept relevant to the role and explore how deeply the candidate can analyze it.

This reveals whether they've memorized surface knowledge or genuinely understand underlying principles. For engineering roles, this might involve discussing material properties or circuit behaviors; for product positions, it might be user interaction patterns. What I'm evaluating isn't specific knowledge but their ability to reason from foundations to applications—a critical skill when working with rapidly evolving technologies.

If you’re curious to learn more about Dystr, you can reach Nabeel on LinkedIn or try their platform at dystr.com

Apprenticeship & Early Career

I’ve always been the kind of kid who took things apart just to see how they worked. School never really clicked for me. I struggled to stay focused and didn’t see myself in a white-collar job. But I thrived in hands-on environments, especially my high school machine shop class. With help from a great shop teacher and guidance counselor, I finished my final credits in a full-time machining co-op program and landed my first job at a local job shop.

I worked my way through a few shops after that, including one with a notoriously tough boss. Those years taught me precision, attention to detail, and what it really means to take pride in your work. I completed my apprenticeship and earned my Machinist Red Seal certification in Canada. That was a big moment for me!

One of the biggest turning points was joining Massiv Die Form, a Canadian manufacturer that builds large cast stamping dies. I finished my apprenticeship there, got hands-on with CNC machine setup, and started to learn programming. I also experienced the kind of camaraderie you only find on a good shop floor. When growth opportunities slowed, I started teaching part-time at the college I’d once attended.

From Massiv to Tesla

A few years later, out of nowhere I got a text from a former colleague at Massiv: “Wanna work at Tesla?” That kicked off my move to Fremont and into the thick of the Model 3 ramp, with what everyone now remembers as the chaos of production hell. I loved it. Fast-paced, high pressure, and between managing the machine shops, I helped launch an internal apprenticeship program with a local college.

Since then, I’ve worked at Apollo Fusion building electric propulsion hardware, stayed through its acquisition by Astra, and joined Neuralink, where I’ve built two shops in two states over the past three years. 

It’s been an incredible journey so far built on grit, adaptability, and a love for building cool stuff with great teams.

Machine Shops

Inside Neuralink’s Machine Shops

As head of the shops, my role is to build environments where engineers and machinists can move fast and work closely together. That means designing spaces that prioritize speed, accuracy, and tight feedback loops, so ideas turn into parts as fast and with as little friction as possible.

Each space has dedicated zones for CNC machining, manual work, and inspection. The workbenches, tooling, and layout were chosen to make it easy for engineers and machinists to work together without friction. Given the kind of hardware we build at Neuralink, the shops are optimized for low-volume, high-mix work: one-off parts, fast revisions, and short lead times.

How Neuralink’s Shops Compare to Tesla’s

Tesla does have internal prototype shops that are similar in spirit to what we’ve built at Neuralink, but I wasn’t working in one of them. I was in the stamping department, and our shop there was more focused on repair work and engineering changes—essentially keeping the production tooling alive and functional. 

Many companies have internal shops, but what matters more than the company is the role the shop plays. Compared to the production shop I worked in at Tesla, Neuralink’s shops are set up considerably differently. They support R&D and early-stage prototyping, which means the layout, tooling, and workflows are optimized for speed and iteration from the start.

Comparing Traditional Job Shops

Most traditional job shops optimize around repeatability, efficiency, and throughput. The business model is built on maximizing spindle time and minimizing changeovers.

Internal shops like the ones at Neuralink (and Tesla) don’t follow that logic. We’re not chasing billable hours. We’re optimizing for tight feedback loops with engineers, IP protection, and proprietary processes. 

Turnaround times are faster because we control the queue and we can pivot in real time. And because we’re embedded with the engineering teams, we don’t just machine parts. We’re part of the design loop, giving a machinist’s feedback to shape the part before it’s finalized.

Day in a Life

Running two internal prototype machine shops in two different states is no joke. Here’s how I’d describe the job if you asked me mid-shift: 

Shop flow: Daily operations and workflow oversight are a must. I’m constantly checking the queues with what’s being machined, what’s due, and what’s coming next. I shift jobs around depending on machine availability, who’s on shift, and where the bottlenecks are. Critical parts get bumped to the top and flagged with the engineers.

Team management: I check in with both shops regularly—sometimes live, sometimes async. Depends on the day. I look for skill gaps, coordinate cross-training, and try to mentor where I can. If something feels off with team dynamics or someone’s performance, I address it early before it snowballs.

Planning and scheduling: We run everything through a custom Jira board. It’s a little hacked-together but works well for our scale. I use it to track open jobs, delivery targets, and status across both shops. It gives me a quick read on where things are slipping and helps me stay ahead of capacity issues.

Technical guidance: I review setups, fixturing, and tool paths when machinists need a second set of eyes. Sometimes it’s just a sanity check before hitting cycle start, other times it’s because parts are failing inspection. If there are machine issues or tooling failures, I work with the team to sort them out fast and keep production moving.

Even though we’re a prototype shop, I try to maintain a high bar on quality and documentation. First-article inspections are a big focus, making sure our parts meet the specified tolerances and requirements. 

Admin and long-term planning: I keep an eye on the usual metrics (on-time delivery, scrap rates, spindle hours) and manage budgets for consumables, tooling, repairs, and CapEx. I’m always looking for ways to streamline: better tooling, smarter setups, or new equipment that’ll save time or improve quality across the board.

Machines & Equipment

What Do You Look for When Buying a New Mill?

When considering a new machine, I start with the basics: Why do I need it? Is it about increasing capacity, adding a new capability, or filling a gap in our current setup?

From there, I look at:

1) Type of machine – 3-axis or 5-axis? Vertical, horizontal, or a mill-turn?
2) Automation – Will it need automation right away, or should I plan for it later?
3) Work envelope – Check X, Y, and Z travel plus table size. It should fit the largest prototype I expect to make, with some buffer.
4) Spindle – Taper and maximum RPM.
5) Tool changer – Make sure it has enough capacity for our mix of jobs.
6) Brand – Reliability, support, and control system familiarity all factor in.
7) Cost – Total cost including tooling and setup.
8) Lead time – I’ve seen quotes with 10-month delivery times, so that’s a real constraint.

What’s the Most Surprisingly Useful Tool You’ve Bought?

I picked up a standard propane tank cage and repurposed it into a locked storage cabinet for sensitive materials. Only the shop team has the passcode!

Background & Experience

Early Career & First Companies

I started out as a software engineer at a SaaS health and safety company in Nashville. Not long after, I moved across the street to a credit card processing company that was in the middle of being acquired. That experience and two colleagues who would later become my co-founders pushed me deeper into startups.

I was obsessed with writing high-quality software, test-driven design, and cybersecurity. So I bootstrapped my first company, focused on helping organizations build a stronger software culture—one where secure, well-tested code was the default, and behavior-driven development kept teams from overengineering.

After a successful exit, I jumped to the opposite end of the startup spectrum launching a venture-funded company from day one. This time, the focus was on harnessing mobile video trends to reinvent how smartphone users told stories through video.

I only realized later how valuable it was to see both sides: a bootstrapped, profit-driven tech company and a venture-funded, growth-at-all-costs one. That contrast shaped how I’d approach future startups: understanding the tradeoff between control over the roadmap and the raw speed that venture funding provides, and how to strike the right balance.

Ventures into Test & Measurement

When I met my co-founder at Initial State, he had just spun some intellectual property out of Lexmark on embedded logic analyzers for custom microprocessors. My background in high-scale cloud computing and his in ASIC design made for an interesting match. As we traded notes, we kept coming back to the same problem—how engineering teams debug complex systems and why so many existing tools fell short.

We eventually scrapped the ASIC IP but held onto what mattered: the core methodologies behind analysis and debugging. Instead of building hardware, we launched a SaaS product designed to make time-series data easy to understand and share. With more devices coming online and systems growing more complex, we saw an opportunity to create tools that helped engineers “see what happened” and communicate it smoothly. As we built out the backbone for data streaming and scaled our enterprise solution, an acquisition offer came our way, setting off a chain of events that led us to Tektronix.

At the time, I was new to the test & measurement industry, but it seemed obvious that things were due for a shift. Ubiquitous connectivity and increasing bandwidth weren’t just making data more accessible—they were unlocking entirely new ways to generate insights in real-time. The kind of data Tektronix had built its legacy on could be used to drive deeper, faster analysis, making problem-solving more intuitive.

I stayed on as we kept pushing forward on this vision, and now being part of a large conglomerate gave me a front-row seat to the sheer complexity of building high-tech products. Along the way, I was introduced to the Japanese concept of “gemba,” the idea that to truly understand a problem, you have to go where the work is happening.

I saw firsthand how engineers wrestled with the vast and often cumbersome tools required to push hardware innovation forward. It was a reminder that the best solutions don’t just solve technical problems; they make the hard parts of engineering a little easier.

Glue

These customer pain points and market shifts led to my latest startup, Glue. It began as an idea we incubated within Tektronix/Fortive, in partnership with their enterprise innovation arm Pioneer Square Labs. Glue’s mission is to transform product realization by giving engineers better tools to measure, validate, and verify designs—bridging the gap between hardware product design and production-ready manufacturing.

Our first product rethinks how test plans and procedures are developed. In product development, engineers start with product designs and specifications (what it should do, and what problems it must solve) to research and development (how it should be done and what’s technically possible). Validation comes next, ensuring the product meets its requirements before it can be produced.

Ensuring a product meets specifications, compliance requirements, and design constraints is often a shared responsibility across many engineering and product teams. Guaranteeing that it performs as expected, under all designed conditions in a repeatable way, typically falls to test and validation engineers. The best teams have specialists dedicated to this, but in reality, many engineers juggle multiple roles. Where this becomes a real challenge is when a product’s complexity or performance characteristics require commensurately complex equipment to acquire signals, measure responses, and test thresholds while also supplying the appropriate loads or supplies to produce real-world integrations.

Setting up and using these instruments is practically a discipline of its own. Dedicated test engineers spend years mastering them, but most engineers don’t have the time. And that’s only part of the problem. The product development timeline assumes there’s plenty of space for testing and validation before launch, ignoring the reality of design changes, rework, and fast iteration cycles. Complex product specs, intricate measurement systems, and rigid validation workflows slow you down at best—and at worst, they lead to overengineering, in-field failures, and missed launch windows.

We automate the entire process of mapping features, requirements, and constraints into structured test plans. Instead of engineers manually figuring out which instruments to use, our system intelligently selects the best tools based on performance, availability, and capability. Test plans that once took hours or days can now be generated in minutes, adjusted instantly, and fine-tuned with both human expertise and machine intelligence. Whenever possible, we automate execution through our deep understanding of programmable interfaces.

Life in a Day

As a Founder and CEO, my role shifts depending on what the company needs most. Some days, I’m talking to investors and studying market trends. Other days, I’m with customers, figuring out where they struggle and how we can help. I spend time with engineers, learning about their latest breakthroughs, and explore new markets where Glue needs a stronger presence. 

A startup forces you to do whatever's necessary, but that’s also the best part—I get to work with great people who turn ideas into real progress. The goal is always the same: build things that make life easier for the people who create. Compared to Tektronix, the biggest difference is speed. The problems change faster, the stakes feel higher, and the work is more varied.

Technical Achievements & Challenges

Some of Glue’s biggest technical breakthroughs have come from orchestrating large language models with advanced knowledge graphs to process massive product specification datasets and validation instrument inventories—all running seamlessly in the background. The result is a simple, intuitive interface for building and refining robust validation and verification plans.

Our biggest challenge right now is developing models that guarantee mathematical correctness and traceability in test plans. At the same time, we’re iterating on cutting-edge models that are constantly evolving. We’re moving fast, but so is AI, and keeping up with that pace is one of the hardest—and most exciting!—parts of the job.

Ideal Hardware Testing Stack

I’ve used a lot of hardware testing instruments, especially during my time at Tektronix. My ideal test setup reflects the same challenges Glue is trying to solve—dealing with a mix of vendors, each with their own automation ecosystems, making integration and usability more complicated than it should be.

For power testing, I like Keysight’s N6705C for programmable DC power and Chroma’s 63206A for DC loads—both excellent, but running on entirely different automation frameworks. Keithley’s DMM6500 or 7510 are my go-to for power analysis and debugging, and while Keithley also makes solid alternatives to Keysight and Chroma, swapping instruments just for better software integration isn’t always practical.

For signal generation and analysis, the Tektronix 6 Series B MSO is my favorite, with the MS05 as a solid all-purpose backup. Tek also makes the best time domain reflectometer in my view. And for quick checks, I always keep a Fluke multimeter and a FLIR thermal camera on hand.

In the end, my bench has about seven instruments from four or five vendors—best-in-class choices, but not built to work seamlessly together. 

Q&A

Top Tech. Innovations You Admire

I might sound like a Tektronix fanboy, but they honestly pioneered some of the most impactful technologies of the last century. Their advancements in the cathode ray tube brought real-time visualizations to monitor electronic signals to see the invisible. That changed everything, opening up new possibilities for electronic hardware.

Favorite Interview Question

“Assume lawmakers in a major metropolitan area implement a new law that requires vehicles to have their headlights on when driving through a construction zone. How would you go about determining the additional energy consumption from this new law?”

I like this type of question because it pushes the interviewee to explore different variables and there’s no perfectly right answer, just a process of refining and discovering what works best.

If you want to learn more about Glue, connect with David on LinkedIn or reach out at david@gluestudio.com

Background & Experience

Academic & Early Career

I was born in Ukraine three months before the Soviet Union collapsed. By the time I was six, we had already gone through three different currencies, and hyperinflation was just wild. Like, the price of dinner could literally change while you were eating it. Store shelves were empty half the time, and people were bartering for basic goods. We moved to the U.S. when I was ten—super unexpected. I didn’t speak any English, and suddenly, I was just here.

Engineering was always in my orbit. My grandfather was an aeronautics engineer—part of the old-school Soviet aerospace tradition. Yuri Gagarin, Soviet engineering feats, all that stuff was a big part of my upbringing. He had this methodical, precise way of thinking about technology, and that really shaped how I approached problem-solving later on.

I was always building things—robots, circuits, whatever I could get my hands on. One of my first real projects was an attempt at an indoor GPS system. I was working with robotics and realized, wait, GPS doesn’t work indoors. That sucked, so I started thinking about how to solve it. My first attempt used ultrasound—I built a system in a physics class that sent sonar pulses and measured the return time. But then I hit a wall because sound doesn’t travel through pockets or solid objects.

I pivoted to radio waves and ended up taking a graduate-level microwave engineering course to figure it out. Eventually, while working at SpaceX, I developed an ultra-wideband RF positioning system with sub-centimeter accuracy—better than traditional GPS. I didn’t commercialize it, which, looking back, was probably a mistake. A few years later, Apple put ultra-wideband chips in iPhones for the same purpose. Timing is everything.

For college, I triple-majored in chemistry, physics, and math. Not because I had some grand plan—more like, I knew I wanted to build things, and I figured the deeper my understanding of fundamental principles, the more adaptable I’d be. At first I was leaning toward research, maybe a PhD, but after graduating, I wanted to check out industry first. I went to the career fair and SpaceX had by far the longest line, and everyone seemed to want in. I ended up spending five years there, working on avionics and power distribution systems.

Building a Startup

I always knew I wasn’t going to stay in one place my whole career. I loved my time at SpaceX, but five years in, I was ready to explore something new. 

After leaving the company, I started consulting. One of those consulting gigs ended up being life-changing. I was helping an investor do due diligence on a sensors company, assessing the technical feasibility of their tech. It turned out he wasn't just evaluating the company, and after working together for a bit he just said:

“You should start a company. I’ll give you a million bucks. Go for it.”

So yeah, I had a blank check to build something—but I didn’t want to jump into just any idea. I spent the next nine months figuring out what problem was actually worth solving.

Exploring Early Ideas

The first ideas weren't about PCB design at first. The investor gave me a lot of leeway to build basically whatever I wanted. So, I spent months researching different industries, looking for a problem where AI and automation could drive real impact.

I kept coming back to my own frustrations with PCB design, especially at SpaceX. Even at one of the most advanced aerospace companies on the planet, PCB layout was still a tedious, manual process. It wasn’t just an annoyance—it was a bottleneck.

I thought back to my first PCB project at SpaceX. The board I designed literally burned up in my hands. I tried using an auto-router, and it was complete garbage. That was my first real "There has to be a better way" moment. And as I kept digging, I realized—traditional PCB design tools haven’t fundamentally changed in decades. AI had disrupted so many other engineering fields, but PCB layout was still stuck in the past.

That’s when it clicked: what if we could build a system that automated PCB layout entirely? Not an "AI co-pilot" that assists a human designer, but something fully autonomous—an actual AI design engine. That’s what led to Quilter.

Quilter

Quilter is an AI-driven PCB layout system that completely automates design. Traditional EDA tools are still manual—you place components, route traces, enforce design rules, all by hand. Some tools have automation features, but they’re half-baked. Quilter does it differently.

We use reinforcement learning and computational physics to generate manufacturable, optimized PCB layouts directly from schematics. No human intervention needed. It’s not trained on past designs or human decisions—just physics and constraints. That means it doesn’t inherit bad habits or biases. It builds boards the right way, from the ground up. We’re already seeing adoption in R&D teams designing IC test boards, evaluation boards, and test fixtures. But this is just the start. AI-driven automation is transforming every industry, and PCB design is no different.

Eventually, designing a board will be as simple as writing a high-level intent, and Quilter will handle the execution.

Landscape of AI PCB Design Tools

Most AI-assisted tools today are "co-pilots" that still require human oversight. They might help with parts of the process—suggesting placements, auto-routing traces, or even optimizing some electrical parameters—but at the end of the day, the bottlenecks are still there.

Quilter is different—it’s fully autonomous, but specifically for PCB layout. We take a schematic that’s already been designed, and we handle placement and routing following your rules and constraints specifications, returning a complete, manufacturing-ready PCB layout.

We focus purely on layout, giving teams an unlimited ability to test design variations without tying up valuable engineering time. Engineers get a fully routed board instantly—faster than a human could do it—allowing them to explore more options, validate concepts earlier, and make better design choices before committing limited PCB resources. This accelerates time to market and enables more innovation while freeing both engineers and PCB designers to focus on refining the optimal board.

It’s also deterministic: Quilter generates predictable, repeatable layouts—so you’re not rolling the dice every time you run it. If you need to make a design change, you can rerun it instantly and see exactly how that impacts the layout, without introducing random variations or unexpected constraints. That level of control means R&D teams can move much faster while staying confident that their designs will work when they hit the lab.

Current Use Cases & Customer Outcomes

We’ve seen some cool use cases. One engineer used Quilter to figure out the minimum viable board size. Instead of manually optimizing, they just submitted the same design at different board sizes—extra small, small, medium, large, extra large. Quilter completed each layout, and they just picked the smallest that worked.

Another big one is treating Quilter like a compiler. In software, you don’t wait until you’ve written every line of code before compiling. You compile constantly—test, debug, iterate. Some of our most advanced users are applying that same philosophy to hardware. They run Quilter multiple times a day on incomplete schematics to check feasibility, catch mistakes early, and explore design variations. That’s huge because it prevents expensive respins down the line.

We’re also seeing SI/PI engineers use Quilter for rapid troubleshooting. Normally, if there’s a signal integrity issue, an SI/PI expert runs a simulation, suggests fixes, and hands it back to a PCB designer, who then has to redo the board. Now, SI/PI experts can just tweak constraints themselves, re-run Quilter, and get an updated layout instantly—cutting out the back-and-forth.

Then, there’s the fun stuff—mechanical keyboards, vintage computers, even a snowman-shaped PCB someone made for holiday decorations.

Long-Term Roadmap

Right now, we’re focused on removing bottlenecks in R&D and test boards. But long-term, Quilter is going to redefine PCB design at scale.

We’re already seeing AI-driven automation disrupt manufacturing, software, and semiconductor design. PCB layout is next. In three years, Quilter will be the industry-standard AI design engine—integrated into MilAero, high-tech manufacturing, and consumer electronics workflows.

The future of PCB design isn’t manual. Engineers will describe what they need, and AI will generate the design. Just like software evolved from low-level assembly to high-level programming, PCB design is heading in the same direction. And we’re the ones making that happen.

Q&A

Top Tech. Innovations You Admire

The MRI, hands down. It’s one of those inventions that sits at the perfect intersection of deep theoretical physics and real-world impact. The fact that we can manipulate quantum spin states to non-invasively see inside the human body is just insane. You’ve got quantum mechanics, electromagnetism, signal processing—it’s all in there. And unlike a lot of physics breakthroughs, which can take decades to translate into something useful, the MRI fundamentally changed medicine in a way that’s still evolving.

Beyond just the science, I love that it actually improves people’s lives. It’s not just a cool piece of tech—it’s enabling early diagnoses, helping doctors catch diseases before they become deadly, and completely changing how we approach healthcare. That’s the kind of engineering I admire: something that’s not just clever, but deeply useful.

Favorite Interview Question

I always ask "why?"—a lot. Not in an annoying way, but to see how deep someone really understands something. The best candidates can explain things at multiple levels. If I ask why enough times and they keep giving me clear, thoughtful answers, I know they actually get it. And if they don’t know? That’s fine—what matters is whether they admit it. The worst thing is when someone tries to BS their way through.

I’m looking for curiosity and honesty. If you’ve never thought about something deeply before, just say so. But if you have thought about it, I want to see how far down the rabbit hole you’ve gone. That’s how I figure out if someone’s the kind of engineer who just follows instructions, or the kind who actually wants to understand how things work at the fundamental level.

If you're interested in learning more about Quilter, stay in touch with Sergiy through email at sergiy@quilter.ai or LinkedIn.

For decades, aerospace and defense have been dominated by traditional prime contractors that built a monolithic hold over the industry. Success was defined by size: the largest budgets, the longest timelines, and increasingly complex projects.

This dominance traces back to WWII, when cost-plus contracts were created to remove financial risk for contractors by reimbursing costs plus a fixed profit. The model spurred rapid scale in emerging sectors like weapons development and manufacturing, especially for R&D-heavy projects. It was a system built for an era of urgent, high-stakes challenges. In the modern era, however, what worked 80 years ago has created unintended consequences. By prioritizing scale and complexity, cost-plus contracts entrenched inefficiencies and rewarded bureaucracy. The post-Cold War era compounded these issues: during the 1990s, a wave of large mergers between defense contractors increased market consolidation and reduced competition with more spending awarded via single-bid solicitations or without competition altogether. At the same time, the continued shift from fixed-price contracts to cost-plus arrangements insulated contractors from risk, further reducing incentives to control costs (1).

The result is an industry stuck in the past where just five contractors control $140 billion annually—about 17% of the $826 billion defense budget.

Recently though, companies like SpaceX and Anduril have started to change the game with a simple idea: speed matters more than size. SpaceX flipped the traditional paradigm of a government-controlled space race to one dominated by private enterprises. Focusing on rapid iteration and reusability, the company cut launch costs by multiple orders of magnitude and enabled frequent access to space. The company proved commercial ventures could tackle massive, high-stakes challenges—and win.

Anduril is applying similar principles to the defense industry, an equally difficult space to break into. Instead of bidding on the Department of Defense request for proposals (RFPs), Anduril directly identified critical defense challenges and took on R&D costs upfront. The company developed systems like autonomous drones for surveillance and reconnaissance, underwater vehicles, and AI-powered software first with the intent to sell directly to the Pentagon later. By reducing execution risk for their customers, Anduril was able to undercut traditional defense contractors on price. This shifted the incentives and pioneered a model where cutting costs and moving fast became the norm.

The landscape for aerospace and defense startups has never been more expansive. The success of these companies has opened the floodgates in venture funding in an industry long thought too slow or complex for startups. A couple data points to illustrate the sheer scale of capital:

• Defense tech startups have raised almost $3B so far in 2024, up from the previous record of $2.6B in 2022. (2)
• Aerospace startups have raised nearly $6B in 2024, after raising a record total of $12.3B in 2021. (3)
• The broader deep tech landscape claims a 20% share of venture capital funding, up from 10% a decade ago. (4)
• Of a16z’s new $7.2B fund, $600M is set aside for companies advancing national interests in areas like space, defense, manufacturing, and robotics.
• Y Combinator backed its first defense startup in the S24 batch (Ares Industries), and has been encouraging defense tech startups to apply since this year.

The future of aerospace and defense will no longer be defined by size. It will be defined by the companies that deliver and move fastest.

While assembling the market map, we researched some sectors that felt like they were born out of science fiction. Some highlights we wanted to mention:

• In-space manufacturing is an emerging sector, with companies like Varda Space manufacturing pharmaceuticals in microgravity.
• There’s significant interest in space-based solar power, with more than nine startups exploring the concept.

Note: We’ve grouped these companies into categories for simplicity, though many could fit in several. Let us know if we’ve missed anything or if you have additional companies to suggest!

In the end, Cadence Design Systems acquired us and we continued developing our tech there. We learned a lot in those 20 months: about the CAD market, what it takes to sell to engineers, and how to build something that attracts an acquirer without needing massive investment. Hopefully, sharing this journey helps others navigate the leap from idea to commercial success.

Background

Meeting my Co-Founders

During my time on campus, I made close friendships with other engineering students, especially through extracurricular teams. My school had a highly competitive Hyperloop team, and in 2017 MUN won 2nd place overall at the annual Hyperloop Pod Competition.

This opened the door to some incredible opportunities. A future co-founder and I were interviewed and brought on as interns at Neuralink, back when the company had fewer than 50 employees.

Late Evenings and Weekend Development

At Neuralink, Darryl and I found ourselves doing the same tedious task over and over: inspecting wire bonds from ASICs on a custom PCB and tracking results in a spreadsheet. The board had four identical ASICs, but each one was rotated differently. As we moved down the array, we would have to “re-map” what each pin did as we looked under magnification. 

After the internship, we headed back to school for our senior year, where we were required to complete a capstone project. The work at Neuralink sparked an idea:

What if we could automate the tedious process of inspecting and mapping these wire bonds?

We originally thought a microscope with an augmented reality interface might boost productivity—or, at the very least, make for a unique project.

Between classes and late nights, we lined up over 50 customer discovery calls with senior engineers to validate our idea. Most agreed: if we could make this “magic microscope,” they’d use it in their labs! With a bit more prodding, they also expressed that they felt a smartphone, laptop + webcam, or even some AR glasses could also work. With a C$500 capstone budget and the need for a prototype, we pivoted from the microscope and into a solution using off-the-shelf (OTS) hardware and software to display CAD data in AR.

By November, we had secured an interview for YC’s W19 batch. This meant round-trip flights for all four of us—over a 10 minute interview!

We booked the flights a week early to make the most of it, and conduct more in-person customer interviews and pitch the idea to other investors in the Bay Area. We didn’t make it into the W19 batch, but one connection from that trip turned out to be pivotal: it led us to our final co-founder and future CEO.

Technical Discussion on Augmented Reality in PCB CAD

There were major components to our solution which would have to be built out in order for this to work, with problems arising with each:

• Problem: File Input. 
  • We had to intelligently interpret a PCB design to display in AR. Gerbers (layer-by-layer image data) were a basic start, though there was much more metadata that we wanted to display.
  • Solution: We found a common manufacturing file input which used an openly available standard and could be exported by all major PCB design tools, IPC2581B.
• Problem: Application clients & augmented reality stack.
  • We needed a client application to execute the computer vision algorithms that displayed information in AR.
  • Solution: We found that we could support most platforms with the Unity game engine, which was becoming an industry standard in augmented reality applications.
• Problem: Backend service to share and distribute information between clients.
  • Solution: There are no mobile PCB design software solutions, however augmented reality software is predominantly run on mobile or glasses based hardware. We built a backend that transmitted design and derivative information between the clients.
• Problem: Frontend service
  • Given the difference in platform technologies between PCB design software and AR software, we also needed a frontend for our backend services to link the two technology stacks together. 
  • Solution: Most PCB design software was built for desktop OS, while our target users were on mobile or glasses-based OS, making a web application the natural choice. This platform eventually evolved into an enterprise portal where teams could be managed for a company’s SaaS plan.

After building our first revision of the software platform, we set out to raise capital to further develop the product and expand our sales efforts.

Investment and Product Market Fit

We met Mihir through a mutual connection—he became our final cofounder and CEO. After a few meetings, our talks had advanced to the point where we knew what we wanted to scale this as a full-time venture.

He saw the vision, knew the customers, and wanted to make a strategic, hands-on investment. It was smart capital. With his background as an EE and experience in his family’s PCB manufacturing business, we decided he would be our best option for the first round. There was a high probability that we would have to pivot our technology towards a manufacturing vertical, away from the engineer who designs the circuit board and more towards the factories that have to build them. Partnering with someone embedded in PCB manufacturing from the start was a big advantage.

Early Hiring and Enterprise Customers

Some of our early issues in sales, product, and eventual solutions can be boiled down as follows:

• Product too early: We kept up with customer relationships and organized pilot projects for the future, once we had the necessary functionality. When needed, we raised additional capital to build out those critical features.
• Limited Capacity for Diligence: We raised capital to make hires dedicated to handling company diligence, including technical hires to develop the product with compliance requirements in mind (SOC2, etc.).
  • It’s best to hold off doing this until you’ve generated enough sales or strong interest to qualify that there exists an addressable market.
• Customer Resistance to Cloud: For clients hesitant to use cloud-based solutions, we developed a self-hosted instance, which also required additional headcount for implementation.

A big turning point for product adoption was at a major user conference for Altium Designer, the leading PCB design software package in our industry. Normally these booths cost quite a bit of money, but we managed to secure a free spot. Our goal for the weekend was to try and have hundreds of interactions with users and get a broad base of feedback for our product.

Talking to our core audience made a big difference: they immediately understood the product's value. Our tool let people working on unfamiliar PCBs save time by reducing context switching between their screen and the device they were testing.

During the conference, we secured numerous highly qualified leads and hundreds more that we were able to reach out to later. Over the next few months, these leads turned into about a dozen trials and a handful of enterprise deals.

This included landing our first enterprise deal with the Philips Sonicare toothbrush team! At first, it seemed a bit surprising to us, but the Philips team turned out to be a great fit for our product. Their smart, IoT toothbrushes and other dental products aligned well with our tool. It was invaluable experience to start navigating sales with large enterprises.

Through this, we also gained hands-on experience working with strict regulations like ITAR and EEA. Handling data subject to these regulations required us to ensure compliance with national security protocols. It was a deep dive into understanding and managing the complexities of regulated environments, that opened up our reach to bigger players in the industry.

We applied a similar playbook to other prospects in our pipeline, and eventually landed more enterprise deals with teams at Google, Fitbit, and Quantel Laser.

COVID-19 and Start of Acquisition Negotiations

In May 2020, we were contacted by the corporate development team of Cadence Design Systems, the largest player in our space. Acquisition had always been the long term strategy of our business and in light of the pandemic, Cadence presented an interesting opportunity to get inspectAR into their customers’ hands while the market was adapting. 

We closed the acquisition in August of 2020 and never looked back! Some key lessons that seem prescient now in hindsight are:

• Leverage co-founders: a productive founding team will be a force multiplier and is exactly what acquirers want from a small strategic acquisition. 
• Focus on becoming a vertical integrator of technologies which are already well established, so you can build a functional product without significant R&D.
• Connect with potential acquirers to build relationships with their customers—or even those of a competitor! We primarily served Altium users at the time of acquisition, which added a strategic advantage.

Background & Experience

Academic & Early Career

I’ve always been drawn to design, but my path to Industrial Design wasn’t straightforward. I attended a pre-college program for Architecture at the Pratt Institute while in high school, expecting it to be my future. Instead, I found myself more intrigued by what my suite mates were doing in Industrial Design. That curiosity led me to dive deep into research, work on my portfolio, and ultimately pursue a degree in Industrial Design at Pratt. 

During my time at Pratt, I explored various aspects of design through internships. I worked in fabrication shops, crafted custom furniture, and even ventured into accessories design at Converse. That part-time role at Converse turned into a full-time position, which was a great start, but I soon found myself itching for something more. 

I reached out to a former professor who had just started her design firm, initially asking her to review my portfolio. To my surprise, she was looking for industrial designers, and I landed the job! It was an incredible opportunity, and I was grateful to have it.

Tangent Design & Starting Makelab

As it happens, the itch returned, and I felt the need to strike out on my own. I began freelancing, taking on clients wherever I could find them. My then-boyfriend, now-husband Manny, and I decided to make it official and co-founded Tangent Design. We had the vision to build our own industrial design consultancy firm, and we were making it work for the first year. During that time, we needed a quick 3D printing solution—I had never used one in my life, and we didn’t own a 3D printer at the time! We found a local manufacturer, 3DUniPrint and they offered great service. We quickly became friends with the owners, another couple.

One year later, they told us they were moving to China with their kids and asked if we would take over their company, and we officially acquired the vendor a couple months later. At first, we saw 3D printing as the stable backbone in consistent revenue for our design firm, but when we tripled our business in the second year, we knew that we had to go all in at Makelab. 

That was back in 2017—seven years ago now!

Makelab

Makelab is an additive manufacturing service, offering parts, prototypes, and production runs for a variety of industries, including robotics, medical, consumer packaged goods, hardware, and manufacturing. We’ve expanded to two locations—our main headquarters in Brooklyn, NY, and a second site in Soma, San Francisco. We serve clients worldwide, working with companies like Peloton, Estee Lauder, Aigen, Lixil, Full Circle Brands, and many more.

We also provide design engineering services, helping companies get their parts ready for printing when their internal teams are stretched thin. Our capabilities range from printing a single part to producing up to 10,000 units.

Unexpected Customer Outcomes

"3D printing might surprise you—yes, it can create reliable watertight seals!"

A great example of how our services add value is our work with Hydrific, a company that develops a smart home water usage sensor. One of their biggest challenges with their Droplet device was creating precise seals and gaskets for prototypes—components that needed to ensure water-tight integrity for field testing. Initially, the Hydrific team didn’t believe a 3D-printed material could meet their needs. They had tried various materials, but nothing seemed to work—everything was either too thick or simply not right for the job. Soft tooling was an option, but it was slow and expensive, which wasn’t ideal for their fast-paced development timeline.

After working with us, they used TEPU 30A (Soft Elastic), a material we offer that’s printed through vision control jetting. It’s a thin, flexible elastomer that can create precise, watertight seals. It allowed Hydrific to produce prototypes faster, cheaper, and with the watertight properties essential for field testing. While solving the immediate challenge, it also opened up new possibilities for future applications like flexible hinges and wearables.

Current Challenges

With a service-based model, you have to grow at the pace of your clients. It’s tempting to make big assumptions and invest heavily in things they might not actually need—mistakes we've seen competitors make time and again. The key lesson here is not to get ahead of yourself. We've learned to really listen to our clients to get it right the first time. That said, a constant challenge with our model is overcoming scaling roadblocks. Clients evolve, use cases evolve, and so do their demands. We’re constantly talking about process innovation and constantly making improvements. It’s important to put together a team that is flexible, agile, and willing to roll with the changes. It’s a part of what I love about this business. It really taps into our design thinking process that I grew to love while studying different types of design in school! 

Life in a Day as a Founder

My focus shifts depending on what we're working on. Some months, I'm deep into process innovation, providing the support our team needs. Other times, I’m focused on marketing, building new revenue streams, or repositioning the business to align with our current operations, offerings, and clients. 

Being a bootstrapped company, I split my time between supporting day-to-day operations and driving advancements and new strategies.

For example, we're currently developing our own software platform, which will give us more control and flexibility over our internal workflow. We're also preparing for our first trade show in September, where I’m heavily involved in the marketing efforts. There's a lot that needs my attention as we get ready for that event.

Q&A

Top Tech. Innovations You Admire

Is it too on-the-nose to say additive manufacturing? It's still a relatively new technology, but the pace of innovation has been incredible. It's given me the opportunity to build an entire company, and I could not be more grateful.

Favorite Interview Question

My favorite question to ask is, "How did you get to today?" I host a podcast called Talking Design and Engineering for the company, and I love it purely for the ability to hear people’s stories. I’ve always been fascinated by the stories behind how someone arrived where they are now-everyone has a story to tell, and I’m always eager to hear it.

If you're interested in learning more about Makelab, stay in touch with Christina on LinkedIn or their website.

Background & Experience

Academic

I grew up in Illinois, where my dad worked as a mechanical engineer for various automotive parts suppliers, designing components like the Mitsubishi Lancer convertible top and the F-150 high-mount stop lamp. Following in his footsteps, I attended the University of Illinois Urbana-Champaign (UIUC) to study Mechanical and Electrical Engineering. Mostly I was just excited to be making stuff with hardware, software, and firmware all together.

Early Career

After graduating, I spent a few years in China at a small startup with a team of seven developing a robot cat toy — not a great business, but an incredibly fun product to work on. My days were filled with coordinating with contract manufacturers, suppliers, and 3PL logistics providers to get our product made, shipped, and in customer hands. Despite the excitement, the majority of my time was spent on operations and I kept thinking:

I wanted to dive into real engineering, but 95% of my time was stuck in program management.

I decided to leave and was fortunate to join Verkada, one of the top hardware startups in Sequoia's portfolio at the time. Between finishing an ongoing build and working through knowledge transfer, I didn’t take a day off between jobs. It was pretty funny—I flew out of Hong Kong, landed at SFO on a Sunday, and went straight to work the next morning!

Startups

At Verkada, I worked on the development of cameras and access control products. However, I quickly realized that the bulk of my time was still spent navigating through spreadsheets. Despite collaborating with some of the top contract manufacturers (CMs) and joint development manufacturers (JDMs) in the industry, most of our efforts focused on aligning our understanding of past, present, and future actions.

After launching several products at Verkada, I wanted to gain more experience in running a company. I joined Athelas as a founding electrical engineer and served as the Head of Product. At the time, we were building a platform for blood diagnostics, IoT remote patient monitoring, and revenue cycle management (RCM) software.

What does medical billing software have to do with hardware?

It turns out that hospitals & providers don’t like doing diagnostics that they can’t easily file claims for, so integrating a robust system was non-negotiable.

The better our RCM product became, the more the providers didn’t care about anything else – in fact, they wanted us to use it to take over all of their medical claims, not just the ones for our products. Why? Because they were used to going line-by-line through medical claims to reconcile dollars and we had built a ton of scaffolding to make that way easier.

It finally clicked: all the problems I had seen in the past, all the “going line-by-line through spreadsheets” weren’t unique to hardware – it was everywhere. Fundamentally, it was a system-of-record and business process management problem. There wasn’t a sufficiently agile and flexible platform for orchestrating workflows between 3rd-party and 1st-party data.

So we went and built Doss.

Experience at Doss

What is Doss Building?

Doss is a lightweight ERP that helps businesses manage their operations from purchase order to point-of-sale.

Our core offering, DossARP (Adaptive Resource Platform), has all of the foundations of a standard ERP: a managed database, configurable workflows and interfaces. The key differentiating factor is that we can get customers up and running in just a few of weeks – sometimes hours. As of today, we’ve rolled out numerous applications across procurement, inventory, order management, invoicing, production planning, and more.

Understanding ERPs & Issues with Current Systems

ERP stands for “Enterprise Resource Planning” which doesn’t really mean anything; except for the “Enterprise” part – that part is true. Companies typically implement these over quarters & years often to the tune of millions of dollars in professional services spend. Oh, and 50% of the time the implementation fails.

Issues with current ERP systems is an easy target to probe, but the reality is that they serve an extremely necessary function. When a business becomes sufficiently large, they need a rigorously defined system-of-record. One that can be imbued with custom business logic, so that standard practices can be adhered to in areas of a business where standard practices are actually a good thing! Invoicing customers, managing treasury accounts, maintaining a ledger of transactions, tracking inventory, processing sales orders, & more.

The problem with enforcing rigor, means that you have to take that complexity and “constrain it” into place so that individual people don’t have to exercise their judgment in any process. This literally means modifying business logic in the actual code of the ERP platform. This can be done by internal specialists if you’re lucky, but mostly it’s outside consultants.

In practice, this means that you’ve just taken a bunch of critical business processes, moved them into an inscrutable black-box, and then you locked it behind a paywall where the gatekeepers are people outside of your company that bill you hourly. Which is obviously a completely insane thing to do.

Doss & "Full-Stack Hardware Companies"

Earlier, I mentioned our "time-to-value," claiming that we can have customers fully operational on end-to-end workflows within a few hours. It might sound too good to be true, but this capability is what sets us apart as a company and makes DossARP perfect for “full-stack hardware” teams managing the complex interplay of hardware, software, manufacturing, procurement, and field service.

Instead of burying business logic deep within ERP code, we elevated it to the application layer, making it fully configurable. Our workflow editor, designed like a flowchart, allows users to easily understand and control specific processes—whether it's handling exceptions or managing data flow to a 3PL or integrated application.

On the surface, our database functions much like a "spreadsheet," with intuitive grouping, pivoting, and filtering capabilities. We even integrated an Excel-like formula editor for creating calculated columns. But it goes deeper with powerful abstractions, such as built-in parent-child relationships, joins, and lookups. For instance, you can effortlessly create a "Line Items" child table for an "Invoice" table, link it to "Payments," and connect everything to "Purchase Orders," enabling easy 2/3-way matching.

Unexpected Customer Outcomes and Current Challenges

If you set out with a goal of (a) deeply understanding a wide collection of businesses and (b) building thoughtful abstractions, you get really interesting results.

Our customers span consumer packaged goods (CPG), food & beverage, apparel, furniture, industrials, aerospace, & more. They’re all unique and distinct if you look at them with 100% fidelity. But if you sand down their edges a bit and squint – they’re mostly the same. I had a hunch this would be the case, but the extent of it surprised me. Across industries, they all manage orders, process quotes, handle payments, store and generate files, communicate via email, and move data between third-party applications.

Just like how humans are 70% water, businesses are 70% PDFs, typed tabular data, and table CRUD (create, read, update, delete).

The question then becomes: what is the greatest common denominator I can build across the patterns we find within customers A, B, and C? And then what about the future, with customer D-through-Z that we haven’t encountered yet? What might they need? Are we sufficiently generalized to solve our near-term use cases and directionally pointed toward our desired long-term capabilities?

The major challenge we’re always facing as a business is mostly about calling our shot and then sequencing the steps. We’re trying to speed-run the capabilities of legacy solutions that have been building since the 1970s. You have to skip a lot of steps and jump straight to the conclusion if you want to catch up.

If you work in aerospace and defense, think of legacy ERP vendors in the same light as prime contractors.

Q&A

Life in a Day

Engaging with prospective customers and gaining a deep understanding of their businesses is the most important part of my work. This often involves either selling or onboarding. The rest of my time is divided among meeting new people to get them excited about Doss, refining product designs, and sharing what we're building with the world.

Top Tech. Innovations You Admire

It's a close tie between the SR-71, the Apollo Program, and the Manhattan Project.

Favorite Interview Question

1. Tell me about something you genuinely love and care about. What do you like about it?

• This question reveals how curious and passionate they are, and how well they can communicate those feelings. If they can’t articulate details about something they truly care about, I question how they'll do the same about their work.

1. If I'm selling to someone: What concerns do you have? You think what we're building is great, you like us, and you understand what we're doing, but what hesitations do you have?"

• This helps me uncover any reservations and allows me to address the core issue directly. It’s also a powerful approach to selling.

If you're interested in learning more about Doss, stay in touch with Wiley through email at wiley@doss.com or LinkedIn.

The Tesla Cybertruck has stirred up controversy in the automotive world long before customers got their hands on it. It defies all conventional car design norms with sharp angles and geometric shapes. The body is made from flat, ultra-hard 30X cold-rolled stainless steel panels. Their decision to build on a unibody car design (body and chassis is a single part) versus a body-on-frame (frame attached to underlying chassis) marks another significant departure. 

For the obvious physical changes above, arguably the most important difference is in the internals of the car: the shift in steering technology. Instead of a traditional mechanical system, the Cybertruck adopted steer-by-wire where pure electronic controls replace the mechanical connection between the steering wheel and car wheels. It’s a complete physical disconnect between the wheel you hold, and the wheels of the car!

In the next sections, we'll dive into a brief history of steering and explore how steer-by-wire actually works (Tesla’ Patent Application: WO2023107555A1). We’ll address how we arrived at steer-by-wire, safety concerns when you remove physical linkages, and implications for future cars.

Steering Technology Evolution

In every moment of driving, one of the primary feedback channels comes from steering. Steering technology has advanced significantly since the early days of mechanical linkages. In the 1900s, early vehicles used a tiller—a simple lever that pivoted the front wheels. While functional, it was limited in precision and control.

Mechanical Steering 

The fundamental components of any mechanical steering system are: 

1. Steering Wheel: Rotational motion is initiated by the driver. 
2. Steering Column: Shaft connects steering wheel to the steering mechanism. When the wheel is turned, the steering column transfers this rotational motion down to the pinion gear.
3. Pinion: Small gear at the end of the steering column. Main function is to convert the rotational motion from the steering column into linear motion. 
4. Rack: Linear gear that meshes with the pinion gear. When the pinion gear turns, it moves the rack left or right. This linear movement of the rack translates into the turning of the vehicle's wheels, allowing the car to change direction.
5. Tie Rod: Located at the end of the rack, it connects to the steering arm. Ensures that the linear motion of the rack is accurately translated into the correct new direction for the wheels.

When the driver turns the wheel, this rotational motion is transferred to the pinion gear through the steering column. The pinion gear then meshes with the rack (linear gear) to adjust the angle of the tires for turning. A tie rod at the end of the rack connects to the steering arm, ensuring this linear motion accurately turns the wheels.

Older cars were often designed with larger steering wheels that provided more mechanical advantage, and a longer throw, meaning each degree of steering wheel turn moved the steering linkage components a greater distance. This made it easier to turn the wheels in pure mechanical systems.

Some car designs use variable ratio steering, where there’s a varying tooth pitch (number of teeth) on the rack changes along its length. In the center, there are fewer teeth per centimeter compared to the ends. This design makes steering more responsive at maximum turn angles and less sensitive near the center, improving maneuverability.

Power Steering

The next major leap came in the 1950s with the introduction of power steering. Hydraulic power steering systems used a pump driven by the engine, to assist the driver in turning the steering wheel. This innovation made steering much easier, particularly at low speeds and parking. By the 1970s, power steering had become a standard feature in most cars.

How it Works: 

1. Torsion Bar (Steering Column): Connects the steering wheel to the pinion gear in the steering column. When the driver turns the steering wheel, the torsion bar twists proportionally to the applied force.
   1. If the force exerted by the driver is insufficient to move the pinion gear, the torsion bar twists, causing a relative displacement between the inner and outer sections of the rotary valve.
2. Valve Operation: This displacement opens specific passages in the rotary valve, directing high-pressure hydraulic fluid to one side of the hydraulic piston.
3. Hydraulic Assist: The pressurized fluid exerts force on the hydraulic piston, which assists in turning the steering gear, thus reducing the manual effort required by the driver.
   1. Degree of assistance provided by the hydraulic system is proportional to the force applied to the steering wheel. Low speed has more assistance, high speed has less. 

TLDR; a hydraulic piston that moves side to side with the steering rack. This video here provides a great explanation. 

Electro-Hydraulic Power Steering

Hydraulic power steering offered better stability and handling, but came with several drawbacks. Since the pump was driven by the engine, there was high parasitic power consumption. At low engine speeds, the assist was weaker and in high-RPM motors, the pump often faced reliability issues.

To fix these problems, electro-hydraulic power steering came into play. Here, an isolated electric motor powers the pump, and a torque sensor measures the force applied at the steering wheel. An electronic control unit (ECU) then figures out how much motor torque should be provided to the hydraulic piston. The addition of an electronic control unit (ECU) to handle sensor inputs meant that the car speed and steering angle could be inputted into more sophisticated control systems. These signals would aggregate to a calculated output motor torque, which would drive the pump and direct high pressure fluids to the desired side of the hydraulic piston.

Electric Power Steering:

Since the late 1980s, electric power steering (EPS) has become the standard in cars. Unlike traditional systems that use a motor to drive a hydraulic pump, EPS connects the motor directly to the steering mechanism.

In rack and pinion setups, the motor can be placed in various positions:

• Steering Column: The motor is mounted on the steering column.
• Pinion: The motor is mounted at the gearbox where the pinion gear contacts the rack.
• Double Pinion: Two pinion gears move the rack. A second rack and pinion, integrated along the rack, has a motor for power assist.
• Steering Rack: The motor is connected directly to the rack, sometimes even concentric with it.

As the industry shifted toward electric power steering, both drivers and manufacturers gained advantages: EPS consumes just a fraction of the energy of traditional hydraulic power steering (~1/20th!), requires less maintenance due to the absence of hydraulic oil, and simplifies vehicle assembly.

Steer-by-Wire

Tesla’s move to eliminate the mechanical connection in the Cybertruck’s steering system is a step evolution forward in automotive technology. About a decade ago, Nissan introduced the first steer-by-wire system in the Infiniti Q50, but still had a mechanical backup linkage for safety. Until recently, regulations required a mechanical link for steering, but changes now allow for systems like Tesla’s. 

This might seem radical especially when it comes to safety, but it aligns with Tesla’s engineering philosophy. Elon Musk’s 5-step approach comes into play when thinking through how their engineering teams rationalize design choices: 

Question every requirement.Delete any part or process you can.Simplify and optimize.Accelerate cycle time.Automate.

How it Works:

The system can be broken down into two primary systems connected through wires instead of the traditional steering column: 

1. Steering Wheel Module 
2. Steering Rack Module 

Steering Wheel Module: This unit consists of the steering wheel, steering wheel angle sensor, and a feedback actuator that provides a haptic road feeling force to the driver. The user imparts a torque, T_h, that changes the angle of the steering wheel, 𝜃_h. This angle, along with the angle of the front tires, 𝜃_f, are sent to the control unit which computes a value for the feedback torque, T_r, to be sent to the driver to reflect road conditions. 

In the Cybertruck’s steering wheel module, a similar set-up is used, but with two separate controllers. In a situation where one controller fails, the system has a secondary controller to provide steering feedback. 

Steering Rack Module: In the rack unit, the road wheel actuator, typically rated between 150 and 1000 W, operates based on commands from the ECU controller. The controller processes inputs from the steering system, along with data from speed sensors, accelerometers, angle sensors, and yaw rate sensors. The motor adjusts the tire angle accordingly, ensuring precise vehicle control. The processed sensor data also generates feedback signals for the steering feedback actuator.

The steering rack module in the Cybertruck’s steer-by-wire system follows the same setup as described earlier, but with a design emphasis on redundancy and fault tolerance. It includes a primary and secondary road wheel actuator, each with its own magnetic position sensor, along with a third inductive position sensor for triple redundancy. This three-sensor architecture enables fault detection and isolation through a two-out-of-three voting mechanism.

The system is also designed with zonal isolation in mind, where the redundant components are physically separated to prevent a common fault from compromising the entire system. This isolation extends across the controllers, power assemblies, and communication networks, each of which is independently managed.

Specifically, the system integrates two zonally isolated motors and controllers for the front road wheels, ensuring that even if one motor or controller fails, the other can maintain steering functionality. The system’s architecture also includes three private communication networks between each node. 

Safety Considerations: 

With no physical link between the steering wheel and the road wheels, Tesla addresses this concern by building redundancy into every critical component. In the Cybertruck, position sensors are triple-redundant, and all other systems are double-redundant. This extends across power supplies, power converters, controllers, and Ethernet & CAN buses. 

Over the course of our research, what we found most surprising is that aviation has been using fly-by-wire systems since the 1970s! It’s important to note that aircraft have mandated inspection requirements by the Federal Aviation Administration (FAA), but cars don't have the same stringent oversight.

Architecture Thoughts:

The primary reason for pursuing steer-by-wire, despite the immense engineering complexity involved in bringing it to mass production, was to enhance steering and handling in large vehicles like the Cybertruck.

Variable Steering Ratios: In traditional systems, the gear ratio between the rack and pinion dictates how much your tires turn when you steer. It’s fixed, no matter the speed. With steer-by-wire, there's no steering column, so the car can calculate the best ratio for any speed—high ratio at low speeds, low ratio at high speeds. Traditional steering wheels usually has a full turn angle of 1080°, but with steer-by-wire, that angle can shrink to 200°.

• In Tesla's Cybertruck, speed-dependent steering ratios enable it to turn with a much smaller radius compared to other trucks of the same length.

Rear-Wheel Steering: Tesla’s steer-by-wire system also utilizes software-controlled rear-wheel steering (as shown below). For comparison, the Ford F-150 Lightning has a turning circle of 47.8 feet, while the Cybertruck with a maximum rear-wheel turn angle of 3°, achieves a turning circle of 43.5 feet. Future software updates aim to increase this angle to around 10°, further reducing the turning circle and enabling even tighter turns.

Steer-by-wire is a glimpse into the future of automotive engineering—simplifying mechanics while unlocking new possibilities for control and handling. Time will tell how widely this steering technology is adopted, but it’s a clear next step in the evolution in steering technology.

Research Links:

• A Composite Control Scheme for Automotive Steer-By-Wire System by Xiaoxue Zhang, Huifang Kong, Hai Wang
• Tesla’s Patent Application: WO2023107555A1