RoboApply

AI candidate-facing job-application app — extracted from the RoboHire monorepo into a standalone, Vercel-deployable system.

• Frontend — Next.js 16 (App Router) at the repo root → deploys to Vercel (roboapply.io).
• API — the Express (TypeScript, ESM) backend under server/, served on Vercel as a single Serverless Function via api/index.ts (export default app).
• Voice worker — interview-agent/ LiveKit voice worker for mock interviews (a separate always-on process; cannot run on Vercel serverless — host it on Render/Railway/Fly).
• Database — Prisma + Neon Postgres, currently shared with RoboHire (see docs/ENGINEERING_PLAN.md for the DB-split path).

See docs/ENGINEERING_PLAN.md for architecture & rationale and docs/TASK_PLAN.md for the phased execution checklist.

Two brands, one shared stack

On the feat/jobright-clone branch this codebase serves two product brands, resolved per request from the Host header: RoboApply (roboapply.io, the international market including Taiwan) and GoApply (goapply.top, mainland China). Both have the same functions and run on the same shared stack: one set of credentials for models, email, voice, storage and web push serves both. A China-specific provider (CN_<NAME> in .env.example) is an optional override of <NAME>, never a prerequisite, so GoApply works with no variable of its own. What differs is what follows from the market: job sources and job-search providers, language, currency and prices, the payment rail (Alipay for GoApply, Stripe for RoboApply), extra sign-in methods and legal lines. Neither brand ever submits a job application for the user. A deployment serves both brands unless ALLOWED_BRANDS narrows it (ALLOWED_BRANDS=roboapply keeps GoApply closed). The plans are indexed in docs/jobright-clone/README.md; the parity rules are in docs/jobright-clone/GOAPPLY_PARITY_PLAN.md. Locally, ./scripts/dev-clone.sh starts both brands at http://localhost:3621 and http://goapply.localhost:3621.

Quick start

cp .env.example .env          # fill in DATABASE_URL, JWT_SECRET, LLM keys, …
npm install                   # installs deps + runs `prisma generate`
npm run dev                   # Express API (.env PORT, default :4607)  +  Next.js :3611 (concurrently)

Open http://localhost:3611. The Next dev server proxies /api/* → the local Express API (NEXT_PUBLIC_API_URL).

To also run the mock-interview voice worker:

cd interview-agent && cp .env.example .env.local   # LiveKit creds
npm install && npm run dev

Scripts

Script	What it does
npm run dev	Express API + Next.js together (dev)
npm run dev:server	Express API only (tsx watch)
npm run dev:web	Next.js only (:3611)
npm run build	The exact Vercel build: prisma generate && tsc -p server && next build
npm run build:server	Compile the Express backend to server/dist
npm run typecheck:server	Type-check the backend (no emit)
npm run db:push	Push the Prisma schema (uses DIRECT_DATABASE_URL)
npm run test	Frontend vitest

Layout

.                       Next.js app (app/, components/, hooks/, lib/, i18n/, …)
├─ api/index.ts         Vercel function entry → re-exports the Express app
├─ server/              Standalone Express backend
│  ├─ src/app.ts        The Express app (roboapply + interview-engine + cron routes)
│  ├─ src/roboapply/    Extracted /api/v1/roboapply/* routers + V2 + engine
│  ├─ src/interview-engine/  Mock-interview control plane
│  ├─ src/cron/handlers.ts   Vercel-Cron HTTP endpoints (CRON_SECRET-gated)
│  ├─ src/{lib,services,agents,middleware,types,brands,utils}/  shared closure
│  └─ prisma/schema.prisma    (generates client to server/src/generated/prisma)
├─ interview-agent/     LiveKit voice worker (deploy separately)
├─ prisma.config.ts     Points prisma at server/prisma/schema.prisma
└─ vercel.json          rewrites (/api/v1/* → the function) + crons + function config

Deploy (Vercel)

1. Import the repo into Vercel (framework auto-detected as Next.js).
2. Enable Fluid Compute (Project → Settings → Functions) and be on Pro (needed for 5+ sub-daily crons and the 300s function duration).
3. Set env vars (see .env.example) — leave NEXT_PUBLIC_API_URL empty in production so the browser calls same-origin /api/v1/*. Set CRON_SECRET, DATABASE_URL (pooled), DIRECT_DATABASE_URL, JWT_SECRET, COOKIE_DOMAIN=.roboapply.io, LLM/Stripe/LiveKit/S3 keys.
4. Deploy. vercel.json already wires the rewrite, the function (maxDuration: 300), and the 7 cron jobs.
5. Host interview-agent/ on an always-on Node host and give it the LiveKit creds + NEXT_PUBLIC_API_URL → the deployed API.

Full deploy runbook + known follow-ups: docs/ENGINEERING_PLAN.md.