photon_interview

Recorded coding-interview app. Two rounds per candidate:

• Round 1 — three questions, 10 min think + 2 min answer each, no retake. Camera + mic + screen all recorded as separate files.
• Round 2 — same questions, take-home, open-ended, re-recordable until submit.

After each upload, webm is transcoded to mp4 in the background (ffmpeg). Admin dashboard at /admin (basic auth).

Stack

Bun · vanilla HTML/JS · ffmpeg · Docker · Railway volume.

Run locally

ADMIN_PASSWORD=changeme bun server.ts

Open http://localhost:3000. Localhost is treated as secure, so camera/screen prompts work.

Deploy on Railway

1. Connect GitHub repo (Dockerfile is auto-detected via railway.json).
2. Attach a Volume mounted at /data (size depends on candidate count — ~1 GB per candidate is a safe estimate at default bitrates).
3. Set variables:
   • ADMIN_PASSWORD — login for /admin
   • SUBMISSIONS_DIR=/data/submissions
4. Generate a public domain. HTTPS is required for camera/screen capture in production.

Layout

server.ts            Bun server: routes, uploads, transcode queue, admin
questions.json       The 3 prompts (edit to change interview content)
public/
  index.html / app.js        Round 1 (timed)
  round2.html / round2.js    Round 2 (take-home)
  admin.html / admin.js      Reviewer dashboard
  style.css
Dockerfile           Installs ffmpeg + Bun runtime
railway.json         Tells Railway to use the Dockerfile

Submissions land in $SUBMISSIONS_DIR/<sessionId>/ with meta.json, resume.<ext>, and round{1,2}/q{1..3}_{camera,screen}.{webm,mp4}.