- RecallMEM - Local-first AI memory built on one rule: the LLM proposes, TypeScript decides. Nothing becomes memory without a verifiable source behind it. No quote, no memory.
- Sprite Agent Workbench - Sprites give an agent a real, persistent computer. A persistent computer accumulates state no chat log shows you. This is the observability layer I wanted, so checking whether an agent is alive doesn’t mean digging through the terminal.
Different apps, same argument: capability isn’t trust. If an AI says it or does it, there has to be a receipt a human can check.
- Speak2Me - Voice AI tutor that teaches through a back-and-forth instead of answering once and going quiet.
- Parallel Lives - AI decision simulator for thinking through possible futures instead of staring at a pros-and-cons list. Same stack powers Journal It (entries become memory, patterns, and eventually your own autobiography) and Speak It (voice-to-text Chrome extension that learns how you talk without storing what you said).
- College Picker - AI college comparison built on real Department of Education data instead of rankings and vibes.
- OriginAI - Chrome extension that checks whether content came from a model, because “sounds human” isn’t proof of anything.
- Upcoming:Data Con LA 2026 (Aug 22, CSU Long Beach) and Commit Your Code 2026 (Sept 3-4, Plano, TX).
- Past:All Things AI 2026, Percona Live (talk recording).
- Cloud Pricing Is A UX Problem - Jun 2026
- Don’t Build A Second Product For Voice - Jun 2026
- Voice Agents Need Receipts - May 2026
- Your AI Agent Is Logged In. That Doesn’t Mean It Should Have Access. - May 2026
- The LLM Was The Easy Part - May 2026
- AI Memory Is Broken Because We Let The Model Own The Truth - May 2026
- Blog - Technical deep-dives into AI apps, memory, voice, infra, and what breaks while building them.
