Add free launch resilience and monitoring

[Rajat77a]

What changed

• Added a local role-aware fallback question bank so interviews can still start during Groq failures or quota limits.
• Added short-lived question caching, request coalescing, shared-cache support, and a Groq circuit breaker.
• Added optional Supabase-backed cross-instance rate limits and shared question caching through supabase-launch-resilience.sql.
• Added privacy-safe structured logs, a public health endpoint, and rate-limited client error reporting.
• Added clear timeout/retry messages without changing the visual design.
• Expanded CI with dependency audit, type checking, linting, production build, and runtime smoke tests.

Why

The free Groq tier is the main launch bottleneck. Previously, identical simultaneous interview starts each consumed a full two-agent workflow, and provider failures could prevent an interview from starting. Serverless in-memory rate limits were also inconsistent across instances.

Impact

Ten identical simultaneous start requests now share one two-agent workflow and receive the same five-question set. Successful sets can be reused briefly, while local fallbacks preserve availability without another paid provider. Shared database controls remain disabled until the included SQL migration is applied.

Validation

• npx tsc --noEmit
• npm run lint
• npm audit --json (0 vulnerabilities)
• npm run build
• Mocked resilience tests for fallback, cache, coalescing, retry, and circuit breaking
• Live Groq test: 10 coalesced callers received five unique questions in 3.28s; evaluation completed in 2.54s

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
prep-peer	Ready	Preview	Aug 10, 2026 11:30am