Birha

AI training data collection platform. Contributors earn money by capturing dashcam video, home environment images, and voice recordings that feed into autonomous vehicle, robotics, and voice AI training pipelines.

Repository Map

Workspace	Path	Stack	Port
Backend API	apps/api	Next.js 15 + React 19	3000
Admin Workbench	apps/admin	Next.js 15 + React 19	3001
Contributor App	apps/mobile	Expo 55 + RN 0.83	Metro
BullMQ Worker	apps/worker	Node.js + BullMQ	—
Annotation Service	apps/annotation	FastAPI + Python	8000
Marketing	apps/web	Static HTML	4173
Static Pages	apps/landing	Static HTML	—
Database	packages/db	Prisma 6 + PostgreSQL	—
Queue Contracts	packages/queue	BullMQ type defs	—
Shared Types	packages/types	TypeScript	—
Shared Utils	packages/utils	TypeScript	—
Shared Config	packages/config	tsconfig + tailwind	—

Prerequisites

• Node.js ≥ 20 (node -v)
• pnpm ≥ 9 (npm i -g pnpm)
• Docker Desktop (PostgreSQL + Redis + MinIO)
• Expo CLI (npm i -g @expo/cli)
• EAS CLI (npm i -g eas-cli)

Quick Start

git clone https://github.com/loxerxxxx/AV && cd AV
pnpm install
npx prisma generate --schema=prisma/schema.prisma   # Required after install — see Prisma note below
cp .env.example .env.local
# Fill: DATABASE_URL, REDIS_URL, SUPABASE_*, S3/MINIO_*, CLERK_*, RAZORPAY_*

docker-compose up -d                                # Start Postgres + Redis + MinIO
npx prisma migrate deploy --schema=prisma/schema.prisma
npx prisma db seed --schema=prisma/schema.prisma    # Seed 15 tasks
pnpm dev                                            # Start all apps in parallel

Key Commands

Command	Description
pnpm dev	Start all apps in parallel via Turborepo
pnpm build	Build all apps (respects dep order)
pnpm turbo typecheck	tsc --noEmit across all 11 workspaces
pnpm db:generate	Regenerate Prisma client after schema changes
pnpm db:seed	Re-seed 15 task definitions
pnpm --filter @birha/api dev	Start only Next.js API (port 3000)
pnpm --filter @birha/admin dev	Start only Admin Workbench (port 3001)
pnpm --filter @birha/worker dev	Start only BullMQ worker
pnpm --filter @birha/mobile dev	Start only Expo Metro
pnpm --filter @birha/annotation dev	Start only FastAPI (port 8000)

Job Pipeline (BullMQ)

apps/api  →  enqueue job  →  Redis  →  apps/worker  →  process

Queue names (from packages/queue/src/index.ts):

Queue	Purpose
birha-process-upload	Post-upload pipeline (validation, S3, metadata)
birha-run-qc	Automated quality control (pHash, GPS, frame analysis)
birha-auto-label	Send batches to FastAPI / Label Studio
fraud-signals	pHash dedup, velocity spike, GPS anomaly checks
payout-processor	Trigger Razorpay payouts
birha-send-notification	Push notifications to contributors

Prisma Note (pnpm + shamefully-hoist)

The root .npmrc sets shamefully-hoist=true. The root prisma/schema.prisma has an explicit output = "../node_modules/@prisma/client/.prisma/client" so TypeScript resolves Prisma generated types correctly.

Required after every pnpm install:

npx prisma generate --schema=prisma/schema.prisma

Health Checks

• GET /api/health — platform health (DB, Redis, queues, auth env, S3, Label Studio)
• bash scripts/smoke-test-e2e.ts — E2E smoke tests
• See docs/AUTH_RUNBOOK.md for auth setup

Store Submission Checklist

• EXPO_PUBLIC_API_URL → production Vercel URL in apps/mobile/eas.json
• EXPO_PUBLIC_CLERK_PUBLISHABLE_KEY = pk_live_* in all EAS production profiles
• prisma migrate deploy against production DB
• prisma db seed against production DB
• Enroll 12 closed testers for Play Store 14-day review period
• GET /api/health returns 200 on production with all checks green
• Google Play service account key at apps/mobile/google-play-service-account.json

Verticals

Code	Domain	Media Type	Typical Task
AV	Autonomous Vehicles	video/mp4, video/quicktime	10-min dashcam loop
ROBOTICS	Home Robotics	image/jpeg, image/png	Object images
VOICE	Voice AI	audio/wav, audio/m4a	Hindi phrase recordings