I build agent evaluation platforms, RL environments and full-stack products. Currently going deep on post-training and physical AI.
| Project | What it is |
|---|---|
| agentforge | Agent bench and grader: Horizon v1 catalog, hidden verifiers, Next.js control plane, isolated evaluator |
| SignalDesk | Local-first GTM workspace: evidence-backed account scoring, traceable workflows, reviewable distribution drafts |
| Primer | Learning app for young children + AI teacher toolkit: 19 live AI tools for worksheets, quizzes and grading, local drills, print/PDF |
| CodeDevAI | Browser IDE that turns a prompt into a running React project: Gemini codegen, saved Convex workspaces, live preview |
| Mock AI | AI mock-interview platform: generated questions, audio or typed answers, speech-to-text, AI feedback and progress tracking |
| PlayGarden | 141 free browser games, no sign-up: custom sizes, speeds and bot strengths |
| QuizSift | Quiz builder: templates, shareable play links, scored results |
| phone-use | Android automation using LLM agents to read screens, tap, swipe and type over ADB; CAPTCHA handoff to humans |
| Patchline | Self-maintaining API clients: compares OpenAPI specs, finds affected code and drafts migration patches |
| Shellcode | Coding-agent shell that routes tasks across models and agent harnesses; tested with a local model and OpenCode runs |