I'm interested in the engineering that makes a model useful: connecting it to the right data and tools, fitting it into an application people actually use, and understanding where it fails.
- Transcribe: desktop dictation for Windows and macOS. Rust, Tauri.
- Paperless-AGX: OCR, semantic search, and answers that cite their pages. Rust, React.
- MatAnyone VCam: real-time background matting as a virtual camera. Rust, TensorRT.
Also working on, mostly in private: coding-agent extensions, semantic code search, and constraint-driven optimization.