I build software that runs AI entirely on the local machine β inference runtimes, the bindings underneath them, and the applications on top.
No Python in the hot path. No cloud round-trip. Nothing leaves the device.
- Isvik β native local inference runtime for Intel AI PCs, exposing OpenAI- and Anthropic-compatible endpoints. 100% Nim + OpenVINO.
- Resonance β the OpenVINO C API binding layer underneath it, with NPU-oriented runtime helpers.
- NimVoice β offline TTS built on Resonance, NPU-accelerated end to end.
- Candlelight β offline Windows endpoint security that uses a local LLM to turn raw alerts into readable threat stories.
| Project | What it is | Stack | License |
|---|---|---|---|
| Isvik | Local inference runtime for Intel AI PCs, OpenAI/Anthropic-compatible API | Nim Β· OpenVINO | Apache-2.0 |
| Resonance | Nim wrapper over the OpenVINO C API + NPU runtime helpers | Nim | Apache-2.0 |
| NimVoice | Fully offline local TTS, Intel NPU accelerated | Nim Β· OpenVINO | MIT |
| Candlelight | Offline Windows antivirus with local-LLM threat narration and a WebView2 UI | C++17 Β· llama.cpp | GPL-3.0 |