Founding Researcher

What we do

The technology used to design hardware is some of the most complex in existence, and it hasn't changed much in decades. Engineers spend enormous amounts of time on setup, scripting, debugging failures, and waiting on runs that take hours or days. A single board program can drag on for 9 to 12 months, mostly because nobody catches a problem with the board until physical testing. By then it's late, and it's expensive to fix.

General-purpose AI and LLMs don't understand circuit physics, so we're building our own models in-house, from the ground up, trained specifically on how signals and fields actually behave on a board. We take schematics straight from the tools engineers already use, run our AI to generate fabrication-ready boards, and hand them back. No rip-and-replace, no throwing away twenty years of muscle memory.

Our vision is simple to say and hard to build: hardware design should move at the speed of software. Chips and boards aren't easy, but the tools around them just haven't kept pace with everything else.

We're a very small technical team by design. We've grown a company from pre-product to nine figures in annual revenue, and grown engineering orgs from 3 people to 50. We've published AI research with real citations behind it and hold multiple patents. We're now hiring the founding researchers on our team — not the fortieth — and you'd be building next to us from day one.

What you'll own

• The core physics-informed AI architectures that learn how signals, power, and EM fields actually behave on a board —* from problem formulation through training to validated accuracy against real hardware.
• The research roadmap: what to model next (signal integrity, thermal, EMI/EMC, manufacturability), what data we need to get there, and how we validate against physical test results, not just benchmarks.
• The data pipelines, simulation environments, and evaluation methodology that let us know a model is actually right before it ever touches a real board.
• The handoff from research to production —* working directly with engineering so what you build ends up generating fabrication-ready boards that engineers trust with real hardware, not research code that never ships.
• The published work and IP that comes out of what we build here —* you'll publish, attend conferences, and represent the science behind the product.
• The research bar and practices for every hire who comes after you. For a while, you're it.

You might thrive here if you...

• Have taken a research idea from open problem to something that shipped and mattered — and want to do it again, this time with the equity, ownership, and title to match what you're actually doing.
• Have deep experience applying ML to physical or engineering systems —* physics-informed models, geometric deep learning, data-driven dynamical systems, or an adjacent domain.
• Have experience with JAX ideally, PyTorch, Tensorflow or similar and scientific computing tools.
• Have a PhD or equivalent research depth in ML, EE, or applied physics is a strong plus, not a hard requirement.
• Have previous experience with numerical simulations for EM and thermal problems is a plus.
• Understand the real trade-offs between model accuracy, training cost, and inference speed, and optimize for what a production system actually needs over the theoretically prettiest solution.
• Want to get fluent, fast, in the engineering discipline —* agentic tooling, CI, production practices —* that turns a research model into something a two-person team can actually ship and maintain, even if that's not your background yet.
• Would rather own one ambiguous, high-stakes research problem than ten well-scoped experiments.
• Are more bothered by indecision than by being wrong —* you'd rather run the experiment, learn, and correct course than deliberate.
  

*human generated em dash

Details

• Full-time
• San Francisco, five days per week onsite
• Reports directly to a founder
• Meaningful founding equity
• Visa sponsorship available