What we build, and how

The apps people actually touch: phone clients that talk to the hardware underneath, and the web consoles the operators behind them live in.

We build the whole path — the phone in a rider's pocket, the API it talks to, and the dashboard an operator watches a fleet from. Because the same team writes the firmware underneath, the protocol between them is designed once rather than negotiated across a vendor boundary.

• Native and cross-platform iOS and Android clients
• React and Next.js web apps, operator consoles, admin tooling
• Bluetooth and realtime telemetry on the client side
• Backend APIs, authentication, payments and billing flows

Low-power firmware for connected devices — BLE and cellular — plus the update pipeline that keeps them current once they're out of reach.

Embedded work is judged years after shipping, on a device nobody can reach. We write firmware for that: conservative power budgets, connections that recover on their own, and an update path that is boring enough to trust. We contribute the parts that belong upstream back to Zephyr.

• Zephyr RTOS and bare-metal C
• Bluetooth Low Energy: GATT services, secure pairing, proximity
• Cellular and GNSS connectivity for devices in the field
• Over-the-air update, provisioning and recovery paths

Electronics from schematic to a board that survives a real vehicle — layout, bring-up, and the test rigs that prove it.

We design the board, bring it up, and build the bench that keeps testing it after we stop looking. That habit is what HardStack came out of — instrumented testbenches that CI and agents can drive remotely.

• Schematic capture and multilayer PCB layout
• Power budgeting for battery-operated devices
• Antenna placement, RF and EMC sanity checks
• Bring-up, instrumented test benches, small-run manufacturing

Bitcoin and EVM work from people who were in it early enough to have written the proofs custodians now publish.

We've built wallets, run nodes, written contracts, and published cryptographic schemes for custodians who need to prove what they hold. The interesting problems here are key handling and verifiability, and that's where we spend the time.

• Bitcoin: wallets, nodes, transaction construction
• EVM smart contracts, integrations and indexing
• Key management and custody design
• Proof of reserves and proof of liabilities

Models put to work inside products — search that understands images, agents that drive real instruments.

We've shipped deep-learning reverse image search at scale, and we now build LLM-backed features and agent workflows — including agents that flash firmware and drive oscilloscopes on a HardStack bench. We care about the evaluation loop as much as the model.

• Computer vision and similarity search
• LLM features, tool use and agent workflows
• Data pipelines, evaluation and monitoring
• Inference at the edge, on constrained devices