I build complete systems, end to end, and I care more about what happens when something fails than about what happens when everything goes right. That shows up the same way whether I'm shipping a production SaaS platform or a distributed runtime for AI agents: understand the real failure mode first, then build for it.
Building in the AI infrastructure space, and taking on full-stack and AI engineering contract work. If the hard part of your project is actually hard, I want to hear about it.
|
UndoLog Crash-safe transaction runtime for AI agent tool calls. Rust, Go, and Python, saga-based rollback, exactly-once execution, 740+ tests. The most thoroughly documented project I've shipped, changelog included. |
Accord Multi-cloud AI agent platform for supply chain reconciliation. Replaced a failure-prone in-memory queue with a Redis Streams architecture, evaluated against Kafka and RabbitMQ, and documented the decision before writing a line of code. |
DevCortex A full-stack developer education platform with true WebSocket real-time chat, not polling. Room-based presence, live typing indicators, 1000+ messages without slowdown. |
Outside of AI infrastructure, I build small, real tools for the Ethiopian market that don't usually get built: a live Telegram bot tracking procurement tenders. Different scale, same instinct, build the thing that's actually useful, not the thing that's easy to demo.