Interview video layer (LiveKit)

The video-conferencing layer for AI-Interviewer, so a one-to-one interview can run in the product instead of on Google Meet. A Next.js client on @livekit/components-react, an Express server that creates interview rooms and issues short-lived LiveKit access tokens, MongoDB for interviews and participants. March 2026, built alone.

Live: https://web-rtc-pied.vercel.app · part of https://satyamkumarsingh.com/work/playground

How it works

dashboard/create-interview  --POST /api/interviews (admin key)-->  Interview + room id in MongoDB
interview/[roomId]          --GET  /api/interviews/:roomId/token-->  LiveKit AccessToken (identity, room grant, TTL)
                            --LiveKit cloud (LIVEKIT_URL)-------->  audio, video, screen share via @livekit/components-react

• server/src/utils/token.ts: AccessToken from livekit-server-sdk with a room-scoped grant; the API key and secret never reach the browser.
• server/src/lib/livekit.ts: RoomServiceClient for room lifecycle.
• server/src/routes/interview.routes.ts: create and list interviews behind authenticate (admin API key); fetch an interview and mint a participant token by room id.
• client/src/app/interview/[roomId]: the room UI on LiveKit's React components; dashboard/create-interview: the admin form.
• render.yaml: the API on Render (listens before the DB connects so health checks pass during cold starts); the client on Vercel.

Run locally

npm run install:all
# server/.env: MONGODB_URI, LIVEKIT_API_KEY, LIVEKIT_API_SECRET, LIVEKIT_URL, CLIENT_URL, ADMIN_API_KEY
npm run dev:server   # :4000
npm run dev:client   # :3000

Status

Working demo on free tiers (Render cold starts apply). The AI-Interviewer product it plugs into lives in a teammate's repository; this repository is the video layer only.