Interview simulation that doesn't let you off easy.

Five personalities. Real time pressure. Honest scoring. Bring your own API key — runs entirely in your browser.

Not a practice tool that tells you you're doing great. A simulation that tells you when you're not.

PrepAIred simulates a real interview — with an interviewer who pushes back, scoring calibrated to be honest, and time pressure that means something.

It is a single HTML file. No server. No account. No data leaving your machine except the API calls you explicitly make to your chosen provider.

🔗 Try it now →

1. Go to ijnebzor.github.io/prepaired
2. Click Get PrepAIred →
3. Choose your provider — Anthropic, OpenAI, or Groq (Groq is free, no credit card)
4. Paste your API key — it validates live against the provider before you can start
5. Watch the Systems Go panel confirm everything is ready
6. Optionally paste your bio or resume for personalised coaching
7. Pick personality, format, difficulty — begin

# Or run locally (API calls blocked from file://, use a server)
git clone https://github.com/ijnebzor/prepaired.git
cd prepaired
python3 -m http.server 8080
# open http://localhost:8080

Keys are validated live before the interview starts, stored in-memory only, and wiped on page unload.

Timer auto-submits when it hits zero. The interviewer notes it.

Each answer is scored 1–10 across four dimensions:

Calibration: 7 = solid. 9 = exceptional. 10 = almost never. Not inflated.

Do-overs are available (max 3 per question) but carry a −0.5 overall penalty per attempt.

• 🔑 Live API key validation — Systems Go panel; Begin button locked until your key passes a live check
• 👤 Candidate profile — paste your bio, resume, or LinkedIn URL for personalised coaching
• 💾 Session export — full session as JSON: questions, scores, verdicts, attempt counts, transcripts
• 🔁 Feedback loop — record the real interview outcome; anonymised locally for Phase 4 calibration
• ↺ Do-overs — up to 3 per question with a score penalty; for genuine blanks, not comfort redos

PrepAIred is built around one principle: your key never leaves your hands.

• API calls go directly from your browser to your chosen provider — no proxy, no backend
• Key stored in JS memory only; wiped on beforeunload
• connect-src CSP restricts outbound to Anthropic, OpenAI, Groq, and Google Fonts only
• No analytics. No telemetry. No third-party scripts.
• No cookies. localStorage stores only anonymised outcome data (capped at 50 entries).
• All user input sanitised before API submission

Known limitations:

• Key readable in browser devtools memory during the session
• AI responses rendered via .innerHTML — a compromised provider endpoint could inject content (mitigated by connect-src CSP)

Full threat model, controls, and responsible disclosure: SECURITY.md

Full detail: phases/ROADMAP.md

Single HTML file. No build process. No dependencies. No framework.

MIT. See SECURITY.md for responsible disclosure.

Built by Benji Zorella · AI-assisted · Brought to life with 🧠 and ❤️ by Claude

"Gold stars are for kindergarten."