Depth over breadth. Two builds I’ll defend in full, then the supporting work. Where something isn’t shipped, it says so.
A permission-aware context layer that turns an organization’s communications into structured, persistent memory — served to both the people and the agents doing the work.
A custom TypeScript MCP orchestrator that runs a self-closing build loop — agents hand off through a shared file with no human in the routing seat.
A Slack-only AI tutor that runs spaced repetition on what you’re learning — SM-2 mastery tracking, no separate app to open. I built it as a multi-agent workflow before it was a tutor.
Deployed and in daily use — by me. Single-user, no public URL, no metrics.
An AI job-application pipeline: source → score → draft ⇄ adversarial-verify → digest, with the submit decision kept human. A self-built artifact of exactly the multi-agent PM work the thesis sells.
(Full architecture lives inside the agentic-build-system case study.)
A platform that paired bootcamp grads into cross-functional teams to ship real apps from PRDs over one-week sprints. Three and a half years, four iterations, the Antler accelerator — closed publicly with accountability when the unit economics didn’t converge.
Real outcome: several team members landed their first industry roles using the work as direct experience.