Hey HN,
I built DiffSwarm, a CLI that orchestrates multiple AI agents in parallel to review pull requests.
The idea is simple: instead of one LLM pass over a diff, spawn 10+ agents that each specialize in a different defect class (concurrency, edge cases,
security, logic errors, etc.). A separate judge agent then cross-verifies candidates — only findings with line-level evidence survive into the final report.
It runs entirely locally. You bring your own Codex or Claude Code credentials, DiffSwarm just orchestrates the processes on your machine. No proxy, no hosted
inference, your code never leaves your box.
I’m a systems-level software engineer with 5+ years of experience in compilers, developer tooling, and AI/AR infrastructure.
Previously at Apple (Clang/Swift compiler team) and SiFive (RISC-V performance and IDE tooling).
Most recently, I was a senior engineer on an AR startup building real-time AI-powered interior design tools using ARKit + RealityKit, with full-stack ML deployment on cloud platforms.
Looking for roles involving AI infrastructure, or high-performance systems where I can have meaningful technical ownership.