I build AI agents that can explain what they did, not just return a plausible answer. I'm especially interested in reliable tool use, temporal memory, grounded retrieval, observability, and the engineering details that make AI systems useful beyond a demo.
I enjoy turning complex agent ideas into clear, inspectable products.
| Project | The idea |
|---|---|
| LinkLoom Python · AI Agent · Temporal memory |
What did we decide, and when did it change? An agent that follows decisions and tasks across meetings, connects their history, and keeps evidence and human review attached to important changes. |
| Real Raise TypeScript · Cloudflare Workers |
Make the numbers make sense. An income and purchasing-power tool that keeps its calculations deterministic and AI explanations optional. |
| Project | Contribution |
|---|---|
| PR-Agent | Merged PR #2589: made files excluded from large-PR reviews visible, so people know what the model did not inspect. |
| Alibaba ANOLISA · AgentSight | 5 merged PRs on agent observability, including TTFT / TPS / TPOT, end-to-end latency, Agent Health percentiles and localization. Explore contributions. |
| Qwen Code | Merged PR #8953: improved Windows file links in session exports and the VS Code companion. |
Agent Runtime Tool Integration RAG Temporal Memory Evaluation Observability




