Sense AI
Founding Software Engineer. Sense AI (a16z), since September 2026.
TypeScript, PostgreSQL, MCP, Streaming STT, TTS
Sense AI is a user-testing platform for coding agents. I shipped its real-time voice interviewer and the MCP server teams use to run their tests.
- Calls placed by the voice interviewer during SF Tech Week
- 291
- Interviews run, 16+ hours, for 21 hackathon teams
- 129
- Tests behind the 15-tool MCP server
- 400+
The interviewer
The interviewer runs streaming speech-to-text, low-latency LLM inference with prompt caching, and text-to-speech that fails over across 3 providers, so one provider going down doesn't end an interview. During a hackathon at SF Tech Week it placed 291 calls and ran 129 interviews, over 16 hours of talking, for 21 teams.
The MCP server
Teams on the platform reach it through a 15-tool MCP server backed by 400+ tests. Bearer tokens are stored as SHA-256 hashes, the account scope comes only from the token, and a test ID that belongs to another account gets rejected.
Never call a tester twice
Dispatch has to survive retries and concurrent workers. Test claims use compare-and-set, and every send reserves a row under a unique constraint, so two workers can't both contact the same tester.
Private uploads
Teams upload their own stimuli. Uploads go through signed URLs, the content type is checked against a 5-type allowlist on the stored object rather than the request, paths are canonical UUIDs, and view links expire after 2 hours. Hence no team can read or overwrite another team's files.
Next: Mend