Product case study / 0066
0.066
Product Engineer · Event flow and AI systems
An in-person AI speed-dating experience with a browser pipeline that turns noisy-room conversations into a printed compatibility report.
Proof signals
- API surface
- 71 endpoints
- Live reach
- About 50 participants across live events
- AI cost
- About $1 per event
User problem
An in-person speed-dating experience needed to capture conversations in a noisy room and turn them into a useful compatibility report.
Ownership
Jaeyong designed the event flow and printed compatibility report, and built the browser pipeline for audio capture and speaker separation.
Selected decisions
- Designed the event flow around an in-person experience and its printed report.
- Used Deepgram to separate speakers in noisy-room audio captured in the browser.
- Routed suitable work to lighter models to keep AI cost at about $1 per event.
Outcome
The system has 71 endpoints, supported about 50 participants across live events, and runs at about $1 in AI cost per event after cost-aware model routing.
Stack
- React
- Next.js
- TypeScript
- Deepgram