ML engineer who ships the whole thing: the model, the backend around it, the infra it runs on and the security review before it goes out.
- 🧠 ML: distilling big LLMs into small local heads, GPU pipelines, benchmarks I can defend
- 🧩 Backend and fullstack: proxies, APIs and the UI on top of them
- ⚙️ DevOps: CI, trusted publishing to PyPI, GPU boxes from a laptop to an AMD MI300X
- 🛡️ Security, blue team: hardening and monitoring, plus redaction and code and dependency scanning in every release
- stuntd: a local proxy that learns your LLM's typed decisions and answers the confident ones with a small head, in about 20 ms instead of a model call. On PyPI, with a Hugging Face Space.
- studiomi300: one prompt to a 30-second reel, with a director agent, a vision critic and image, video, music and voice models on a single AMD Instinct MI300X. Built solo for the AMD Developer Hackathon, May 2026: finalist, winner of the Hugging Face track and the Qwen Special Prize.
- Finalist of the AMD Developer Hackathon by lablab.ai, May 4 to 10, 2026, and winner of the Hugging Face track and the Qwen Special Prize with studiomi300, built on AMD ROCm and the AMD Developer Cloud (certificate)
- Laya: fixed its packaging (Python floor and a stale metadata copy); stuntd is listed in its Community Tools
- stuntd is listed in awesome-jev and four other Jev lists
