Seven years building reliable distributed systems and release infrastructure. Now architecting an outer/inner-loop multi-agent platform on the Anthropic MCP protocol — with the boring rigor (typed pipelines, deterministic guardrails, observable rollouts) that keeps AI from being a demo.
Multi-agent orchestration for release readiness. Outer-loop principals (Route · Check · Fetch · Scan · Build) govern policy and learn across engagements; inner-loop execution agents run inside client projects. Step-based pipeline with typed status returns and roll-up resolution.
An open-source MCP server that answers a single question: is this ready to ship? Four tools — analyze_app, explore_auth, detect_auth, qulib_score_automation — with auth-aware scanning and a deterministic release-confidence score.
A LeetCode-style platform for QA engineers — a deliberately broken app with 48 documented bugs across UI, API, events, accessibility, performance, and mobile. Dual auth, challenge engine, severity-weighted scoring.
Below is a runnable model of the orchestration layer I'm building — five principal agents executing a release-readiness sweep against a target deployment. Press run to dispatch.
Working on AI infrastructure, agentic platforms, or release-engineering at scale? Send a note. I reply to most things within a day.