Something impressive you've built or achieved?

One of the most technically demanding things I've built is a custom handwired, split mechanical keyboard, a variant of the dactyl manuform, designed and fabricated almost entirely from scratch. I modeled the case in Fusion 360 and 3D printed it on my own printer, then CNC'ed a clear acrylic base plate so the internals, the diodes, switches, wiring, battery, etc., are fully visible. Each half runs its own nRF52840 microcontroller; the two communicate wirelessly over Bluetooth, with one side relaying input to the host computer. I used ZMK firmware which handled a lot of the communication and reliable low-latency pairing between the two halves, but it still required a lot of configuration. For power, I ran a power profile to estimate real-world draw and sized each half with a 2000mAh battery, cheap and small enough to fit the case without adding bulk, which ends up lasting over 6 months on a single charge. It's now my default travel keyboard.

A time you tackled a problem in a novel way and achieved something truly impressive.

I spent 2 years on the Amazon Glow team (an interactive tabletop video-calling and projection device for children), joining when it was a 30-person team and staying through launch. I worked on projects spanning UX, software, and hardware engineering.

The product team was focused on digital-only experiences. Apps were essentially synchronized activities layered on video calling, like reading a digital book together where both sides could flip pages. I saw an opportunity the team hadn't explored: using the device's RGB and IR cameras to detect physical objects, and its projector to bring them to life in the shared play space. I built prototypes and wrote an opportunity paper detailing them, along with the hardware limitations - processor speed, projector latency, interference - and realistic constraints on how responsive this kind of interaction could actually be. Based on that paper, leadership changed product direction, scrapping several planned apps in favor of these differentiated, physical-object-based activities.

I was then asked to lead the technical and UX capability workstream, working across CVML, software, hardware, UX, and industrial design. The central challenge was designing a fiducial marker system (the pattern the camera uses to identify objects) that satisfied three competing constraints: it had to look abstract and natural enough not to be visually distracting (unlike a typical QR-style fiducial), stay reliably detectable under projector interference, and run through a detection algorithm fast enough for the hardware's limited processor. We solved all three, and the resulting design shipped in Glow's first-party app lineup, including tangrams and a puzzle game, as the first apps beyond the originally-approved drawing and reading experiences to involve physical objects. Plus we got a patent application out of it.

What makes you an outlier?

What makes me an outlier is technical breadth combined with a specific approach to learning: I believe any domain can be picked up quickly if you break it into digestible parts and know when to stop. Pushing for more depth than a goal requires is often just diminishing returns.

I spent 6 years in Amazon's Concept Lab investigating new hardware and technology capabilities, which meant constantly learning domains from scratch and building customer experiences before the tooling to make them easy existed. I built apps spanning AR/VR/3D and a range of hardware (IoT, RFID, other location-detection systems), each requiring me to get functionally deep in an unfamiliar stack fast enough to ship something real. That work resulted in over 25 patents, on most of which I'm the primary inventor.

The same pattern shows up outside of work: I've done CAD design, CNC fabrication, and embedded RF firmware to build a custom keyboard from scratch, and used community organizing plus an engineering constraint I measured myself to change how an 87-unit HOA allocated parking. None of these are things I studied formally. I learned each one just deep enough to build something real and make a defensible decision.

The payoff is pattern-matching across fields: a constraint or design pattern from one discipline (mechanical tolerances, RF power budgets, hardware detection limits, software architecture, user research strategies) often reframes a problem in a completely different domain faster than staying within one field would.

What have you done that's exceptional?

I have investments spread across 5 brokerage accounts, which made it hard to get a real picture of how my portfolio was performing. I'd learned how to build agents and tools recently, so instead of using an aggregator, I built a personal tool: I could upload account statement PDFs and it would parse and organize everything into my own database, then let me ask it questions like a financial advisor - how to rebalance, what my actual risk exposure looked like across accounts. The insights it gave were genuinely useful.

When I showed it to my wife, she said: "would be cool if it just traded stocks for you." That offhand comment turned into a real feature. I built an LLM-driven trading agent. It pulls live positions, technical indicators, and news, reasons through trade decisions, and passes every signal through a risk-validation layer before anything executes through Alpaca. I ran dozens of backtests across different models and configurations; the results were promising but not close to statistically significant. Full-year backtests are expensive to run, and I didn't need proof, I needed enough signal to justify moving forward. So I moved to a paper trading account and let it run for real. A month in, it's beating the S&P by about 2%, which means nothing statistically yet, but the system is live and I'm tracking it seriously.

What's stuck with me most isn't the returns, it's how much the process taught me about how LLMs actually reason under uncertainty. I'm not a trading expert; the friction of manually researching and placing every trade is what kept me from investing more actively in the first place. Watching the model's reasoning, especially the vocabulary it uses when it reads news and weighs a decision, has taught me more about both markets and how these models think than anything I'd read about either topic.

What's something weird or niche you were obsessed with?

Mechanical keyboards. I own around 50, roughly 10 are fully custom, and I've made and given away a good number as gifts. And by "made," I don't mean assembling a kit: I create custom CAD designs, then 3D print or CNC mill the case and laser cut the switch plates myself. I love the tactility, the functionality, the customizability. I wasn't even a touch typist until about a decade ago, when I got my first mechanical keyboard, and from there it became an obsession with optimizing sound, tactility, ergonomics, and typing speed. I went from fully-assembled boards, to building my own from kits and custom switches, to eventually modeling my own in Fusion 360 and fabricating them myself. Usually 3D printed, though some of my favorites are milled from solid walnut. This obsession ended up being a catalyst for a bunch of other hobbies and skills I probably never would have picked up otherwise: CAD design, 3D printing, laser cutting and CNC milling, microcontroller programming, and handwiring electronics.

A time you successfully hacked a non-computer system.

A few years ago, I was HOA president of an 87-unit community. I didn't have unilateral authority to reallocate our narrow private road, and parking disputes are notoriously political. Rather than spend months lobbying the board directly, I went to the fire department first. Since the road was private, we technically controlled it, but I needed their read on the actual safety constraint. It turned out to be simple. Could their trucks still get through if we added parking on one side? I measured the truck's width against the lane, worked with the department to confirm the numbers, and brought their explicit sign-off to the board. What had been an unresolvable "is this safe" argument became a settled fact, and the board approved a plan that increased capacity by nearly 50%.