I build research tools, public records projects, and practical software for work that needs more proof and less theater.
The thread through my work is simple: name the source, state the limits, and leave a record other people can check.
| Project | What it does |
|---|---|
| AI Writing Skill Field Guide | Measures the maintenance of 68 open-source AI-writing skills, publishes the data and method, and reports the negative results alongside the successful ones. |
| email-agent | Teaches the agent loop through a deliberately small email-triage example, with the shortcuts and safety boundaries called out in the code. |
| YouTube Transcribe Rebuild | A macOS-first Tauri app that uses yt-dlp, ffmpeg, and whisper.cpp to turn YouTube videos or local media into local TXT and SRT transcripts. |
| Receipts Not Vibes | A dependency-free landing page that turns measured findings into a literal receipt and keeps the failed experiment visible. |
My projects favor primary records, reproducible methods, and plain statements about what the evidence cannot prove. That matters most when institutions hold more power than the people described in their records.
I also build to learn. Earlier work includes a Java media player, a data structures course repository, and BookBot. I keep those projects public because the path into the work belongs beside the polished results.



