Self-taught Linux engineer working close to the hardware — kernel modules, systemd, udev, core-dump debugging — currently focused on running AI locally on AMD Ryzen AI NPU silicon, and on making that stack reproducible.
What I've built
- npu-container — the whole NPU inference stack in a container: digest-pinned reproducible build, a preflight/verification harness, GitHub Actions CI, and Kubernetes manifests.
- local-ai-assistant — a fully offline voice assistant on the NPU: mic → Whisper ASR → on-NPU LLM → piper TTS. Stdlib-only Python.
- npu-whisper — one-shot Whisper transcription on the NPU (Bash, zero dependencies) plus a RAPL power/latency benchmark harness. Measured RTF ≈ 0.17 and ~10× lower energy per transcription than the same model on CPU.
Write-up: Running Whisper + LLMs on an AMD NPU under Linux
— the amdxdna / XRT driver stack, the one real gotcha (memlock), the runtime
that skips the Windows-only SDK, and benchmarks you can reproduce.
Stack: Linux internals · Python · Bash · Ryzen AI / XDNA2 · FastFlowLM · Docker · GitHub Actions · systemd / udev / DKMS · Git
Learning: Kubernetes in anger · Terraform · AWS
Open to junior MLOps / DevOps / SRE / Linux / backend-Python roles. 📫 jac-76@live.com · 🌐 jac-76.github.io