Product I founded
Lobbyee
AI training simulator for hospitality front-line staff. Practise hard guest conversations against a realtime voice AI, with transcript-grounded coaching
What I built
- Built a realtime voice pipeline on pipecat and WebRTC, fixing a production failure caused by TURN relay selection
- Designed a multi-tenant Postgres schema with row-level security, verified by adversarial cross-tenant tests
- Maintained a 946-test suite behind six CI workflows, including 173 tenant-isolation tests and a weekly restore drill
- Built a gold-set evaluation harness for LLM grading that caught a token-truncation bug corrupting scored reports
- Shipped OTP auth, pooled subscription billing and transactional email with error monitoring and server-side analytics
How the numbers were measured
Maintained a 946-test suite behind six CI workflows, including 173 tenant-isolation tests and a weekly restore drill
Read off the CI run for mayankgoel214/lobbyee on 2026-08-27 (run 33075822961): the Typecheck/Lint/Test/Build job reports 946 passed and 175 skipped across 97 files, and the separate "Tenant isolation (RLS). HARD GATE" job reports 173 passed. The three jobs are `quality`, `secret-scan` and `tenant-isolation` in .github/workflows/ci.yml.
Built with
Next.jsTypeScriptSupabasePostgreSQLPrismaVercelPipecatWebRTCPythonSentryPostHogTailwind