Real systems — mostly in production, honestly labeled when not. A running thread of retrieval, multimodal, and AI-tooling work built to survive real use.
A production search engine that indexes podcast and educational video at the frame and transcript level, so users find the exact moment something was said — by meaning, not keywords.
Raw speech transcribed and indexed semantically, so a rambling idea spoken weeks ago becomes searchable by association, not exact words.
A shared memory layer that indexes decisions and context as teams work, so knowledge is retrievable later instead of lost in someone's head or a Slack thread.
A full mobile-to-backend system where a photo is the only input: vision models identify the meal and estimate calories and macros automatically.
Reusable context docs that encode a team's real conventions, so an AI coding tool follows them automatically instead of being re-prompted into compliance on every task.