Skip to content

Work

Backend systems, agents, and the evaluation harnesses behind them.

  • An agentic RAG system for macOS troubleshooting — a branching LangGraph, a domain knowledge base built from 9 scrapers, and a blind-judge evaluation harness that blocks its own deploys on quality regressions.

    · Solo — design, build, deploy, evaluate
    • langgraph
    • agentic-rag
    • evaluation
    • fastapi
    • python
    Live site
  • A coding agent built on LangGraph from scratch — a planner, a ReAct tool loop, and a network-disabled Docker sandbox — graded against 50 real Django and Sphinx issues from SWE-bench Verified Mini.

    · Solo — design, build, benchmark
    • langgraph
    • agents
    • evaluation
    • swe-bench
    • python