I work on the plant floor and build systems and tools on nights and weekends. What I build changes, but the approach stays the same: trace the failure, plan the next move, and keep eyes on what the system is actually doing.
LangGraph, RAG, evals, and output guards. Agents that fail loudly, log honestly, and stay debuggable.
Supply Chain Multi-Agent SystemStart at the failed output and walk back through every condition until the cause is in hand. Years of doing this on real PLCs taught me to make the LLM narrate what a deterministic walker already found, so it never reasons about ladder semantics from scratch.
Break the work into shippable milestones with typed state and clear gates. Real-LLM integration tests run in CI, so a regression shows up before merge instead of after.
Output guards with deterministic rules first and an LLM judge second. Self-hosted Langfuse for traces. Nothing ships as a black box.
Shipped systems and tools across Rust, Python, Kotlin, and one production API written entirely in COBOL.
A from-scratch Rust BitTorrent engine, published to crates.io and daily-driven as my torrent client.
One Rust core behind a desktop app, a web UI, and a CLI. A single-binary Sonarr and Radarr replacement.
Google Inbox, rebuilt as a native Rust desktop email client I use every day.
A production HTTP/JSON API written entirely in COBOL, live on the public internet.
Three specialist agents coordinating through a typed LangGraph state machine
AI-powered Allen-Bradley PLC ladder logic troubleshooter
A terminal emulator built from the core up in Rust. An honest, in-progress build.
Offline-first Android debt tracker, published on Google Play
Single-binary Rust service bridging Telegram to Claude Code sessions