independent systems engineer
Multi-camera vision, on-device language models, and sandboxed edge runtimes, all running on my own GPUs and hardware. One command up, and I'd rather show you a live demo than tell you it works.
01
Self-hosted, multi-camera AI security platform.
Person, face, and vehicle detection plus a 19-tool LLM chat assistant over live RTSP feeds, with Telegram alerts (AI scene descriptions) and live Grafana/Prometheus metrics. All local via Docker Compose, zero cloud.
Built and tested on a dual-GPU rig (RTX 5070 Ti + RTX 3090); tiers for smaller cards.
02
A web UI for NVIDIA LocateAnything-3B.
One prompt, six grounding tasks on your own GPU: detection, phrase grounding, OCR, document layout, GUI grounding, and pointing. Describe what to find in plain English and get boxes back. One `docker compose up`, with a no-GPU mock mode for development.
UI code is Apache-2.0; the NVIDIA LocateAnything-3B model is under a non-commercial research license.
03
Hands-on industrial control-systems security rig.
A hub-and-spoke testbed on real industrial gear: a Siemens S7-1200 PLC, Kunbus RevPi, Raspberry Pi, and Arduino, wired into IEC 62443 security zones with a live sensor→PLC→dashboard pipeline you can watch react in real time.
Real PLCs, relay banks, and buck-converter power stages, documented end to end, down to a hand-drawn star-ground schematic.
04
Privacy-first nursery sound monitor.
Listens for sustained noise at the nursery door and ducks the living-room TV in seconds, detects crying via spectral analysis, and pings my phone. Runs entirely on the home LAN, with no cloud and no audio recording.
Raspberry Pi Pico 2 W with an I²S mic, controlling the TV over the LAN.
Hot-swappable sandboxed plugins on a Pi.
A secure IoT runtime that runs untrusted Python plugins on Raspberry Pi via the WASI Component Model. Hot-swap a running driver in under 10ms.
Distributed consensus in the browser and on a Pi cluster.
The same Rust binary runs in the browser and on a Raspberry Pi cluster: leader election, log replication, and partitions visualised live. 120+ tests, chaos controls.
Production-style Cloudflare Workers + Leptos.
URL shortener (Workers KV), an edge rate limiter, and a capability sandbox. The same capability-security ideas as WASI, at the cloud edge.
Image / audio / video conversion, 100% in-browser.
Batch-convert media with FFmpeg.wasm. Nothing is uploaded, and files never leave the device.
An interactive Rust → WASM terminal (for fun).
My previous portfolio: a fully interactive terminal compiled from Rust to WebAssembly, with working commands and zero JavaScript.
I'm an independent engineer, and I build complete systems end to end: the hardware, the backend, the frontend, and the CI that keeps them honest.
I lean self-hosted, zero-cloud, and privacy-first. I like things that run on your own machines and don't phone home, and I'd rather show real numbers than make a claim.
Lately that's meant self-hosted AI: multi-camera vision, on-device LLMs, and open-vocabulary detection, sitting on top of years of edge, embedded, and distributed-systems work.
Before that, I spent a few years building web3 and crypto products, including co-founding CyberVerse, with tooling across Ergo, Solana, and Cardano.