Building
Mozart AI.
I worked across product and engineering: core creation surfaces, collaboration, onboarding, model evaluation, and the analytics used to decide what came next.
A Library built around songs
I took a songs-first Library redesign from product requirements to production, bringing generated music, alternate takes, playback, and editing into one surface. I owned the interface, state model, and service integration.
A studio you can share
I built real-time collaboration with authenticated rooms, durable session state, live tracks, mixer controls, MIDI, and collaborator presence. I also built the onboarding for joining and starting a shared session.
Learning by doing
I built onboarding that advances when a user performs the real gesture. Conditional steps, surface restoration, and event instrumentation made the tour measurable and resilient.
Model evaluation for music
I developed a TypeScript audio-evaluation harness around a 227-clip corpus, deterministic BPM, key, and lyric checks, and human A/B listening. I also worked on a LiveKit voice pipeline for music sessions, where voice activity and turn-taking have to contend with music in the room.
Evidence after the release
I defined typed PostHog event contracts and dashboards for creation, collaboration, and export. Funnel and cohort analysis identified friction and informed product decisions.