I like building software around the places where systems become hard to reason about: concurrent changes, durable execution, real-time collaboration, state, and failure.
The goal is usually the same — make the hidden behavior visible, reproducible, and useful to an engineer.
|
When branches pass alone and fail together. An integration acceptance gate for concurrent changes. ProofMerge assembles candidate branches in an isolated worktree, verifies the combined application, and returns
|
Agent workflows should be replayable, traceable, and governed. A durable runtime for multi-agent LLM workflows built around deterministic execution, OpenTelemetry traces, capability-scoped tools, evaluation, and replay.
|
|
Collaborative coding without the tab-switching tax. Real-time coding rooms with shared Monaco editing, Yjs synchronization, Socket.IO, WebRTC video, Judge0 execution, and multi-language problem solving.
|
Market mechanics instead of another CRUD demo. A full-stack exchange simulator with a price-time matching engine, simulated wallets, double-entry accounting, WebSocket updates, and live reference charts.
|
reproducible failure > mysterious failure
evidence > assumptions
small interfaces > framework magic
explicit state > invisible state
boring reliability > clever fragility
I am most interested in the point where a useful demo has to become a system you can inspect, replay, debug, and trust.
developer infrastructure
distributed systems
AI reliability
runtime design
observability
collaborative software
Building things that are useful to inspect, reason about, and trust.