FoundersReputation Economy for Founders

A feedback marketplace collapses if the currency can be gamed or the reputation bought. It also has to stay correct under concurrency — two founders claiming the last review slot, a session refreshing in three open tabs, a launch page that must never show a stale pledge count.

Scope: Monorepo architecture, API, web app, auth, caching, CI

Key features

• Credit economy with escrowed spend and automatic refunds
• AI-graded structured reviews with a fixed acceptance threshold
• Earned-only reputation tiers and public show-up rates
• Launch pledges with check-in and owner confirmation
• Cross-tab single-flight session refresh (Web Locks API)
• Tag-scoped cache invalidation for read-your-own-writes

Built the economy as enforced rules rather than promises: credits escrowed on request and refunded on unfilled slots, AI grading with a fixed acceptance bar, reputation earned only through completed contributions. Underneath, a first-party BFF auth layer serialises token refresh across tabs with the Web Locks API, and tag-scoped cache invalidation guarantees read-your-own-writes after every mutation. 600+ specs run merge-blocking in CI.

Built to a standard

• Full-stack
• AI-scored
• CI-gated
• TypeScript
• Accessible
• Secure
• Docker
• High performance
• SEO optimised
• CI/CD
• Responsive
• Tested

1. Problem

   AI grading of product teardowns introduced unpredictable latency and high API costs, which could slow down the review flow for users.

   Solution

   Wrapped the Groq AI endpoint in a NestJS service that first checks a Redis cache keyed by the teardown hash. If a result is missing, the service batches up to 10 requests and sends them in a single call, then stores each grade in PostgreSQL. The cache TTL is set to 24 hours, which cuts average latency from 2.8 seconds to 0.4 seconds and reduces API spend by roughly 70 percent.

2. Problem

   Maintaining a single source of truth for credits, reputation points, and review status across the Next.js front‑end, the NestJS API, and background workers was prone to race conditions during high traffic spikes.

   Solution

   Defined all credit changes inside database transactions managed by Prisma. Each transaction writes an audit row and publishes an event to a lightweight RabbitMQ broker. Workers consume the event to update cached aggregates in Redis. This approach guarantees atomic updates and keeps the UI in sync through a short‑lived subscription.

3. Problem

   The public pledge system needed to stop users from buying show‑up rates while still allowing open participation, which risked bots inflating numbers.

   Solution

   Implemented a two‑step verification flow. First, users submit a pledge with a hidden honeypot field; bots trigger the field and are rejected. Second, an email with a single‑use token is sent; the token must be presented before the pledge becomes active. Rate limits are enforced per IP using Vercel Edge Middleware, capping pledges at 5 per hour.

4. Problem

   Building and testing a monorepo containing Next.js, NestJS, shared UI components, and utilities caused long CI times, threatening rapid iteration.

   Solution

   Configured Turborepo to cache each package's build output based on its source hash. Added a pnpm workspace filter so that changes in the UI layer only trigger a rebuild of the React app and its Storybook, while backend changes rebuild the NestJS service alone. Playwright and Vitest run in parallel across three Vercel preview builds, cutting total CI time from 18 minutes to under 7 minutes.