Engineering leader and hands-on builder. I ship production software across the stack — systems-level C, Go services, Python AI pipelines, and TypeScript front-ends — with a bias for local-first, no-SaaS tools that run on your own machine.
Most of what's below is built end-to-end: the hard backend, the GPU kernel, and the README.
| Project | What it is | Stack |
|---|---|---|
| pdfcracker | Recover passwords from encrypted PDFs on macOS — all encryption revisions (R2–R6), 15+ attack modes, GPU acceleration via Metal, ARM NEON SIMD, distributed cracking. Zero dependencies. | C Metal SIMD |
| imageclust | Clusters photos by what they're about, not just how they look — CLIP ViT-L/14 embeddings + Ward hierarchical clustering + Ollama-generated labels. Entirely on-device. | Go React CLIP |
| corpus | Ask natural-language questions over your own notes, PDFs, and docs. Hybrid semantic+keyword search, multi-hop reference expansion, 7-tool MCP server for Claude Code. pip install corpus-rag |
Python RAG MCP |
| cleancut | Drop in a video, get back a cleaned .mp4 — profanity muted, explicit scenes cut. Layered local AI stack: Whisper, Ollama, NudeNet + LLaVA, HF audio models. Every cut is auditable. |
Python Whisper Vision |
| TPSGenerator | Java load tester for HTTP APIs — 12+ traffic patterns, lock-free HdrHistogram metrics, circuit breaker, real-time resource monitoring. Pairs with TPSGenerator-Server. | Java Concurrency |
| claude-code-skills | 12 drop-in Claude Code skills for dev workflow and AWS ops — auto-discover your AWS resources at runtime, no config files. | Tooling AWS DX |
Languages Go · C · Python · TypeScript · Java
AI / ML Local-first inference (CLIP, Whisper, Ollama) · RAG · MCP
Systems Metal GPU compute · ARM NEON SIMD · lock-free concurrency · AWS serverless
Most projects run entirely on-device — no cloud APIs, no data leaving your machine. I care about software people can actually own and audit.