Hey Product Hunt 👋 I kept freezing in technical interviews even when I knew the material — the gap was never knowledge, it was reps. Paid mock- interview services were expensive, and the "AI" ones mostly wrap ChatGPT, so your answers get shipped off to a third party. I wanted something I could drill with cheaply and privately. So I built Peakblick: you paste a job description, it interviews you like a real technical round — one question at a time, on a timer — and a second model scores every answer with written feedback on what was missing. The part that took longest: instead of wrapping GPT, I trained a 1B model from scratch to do the scoring. Two reasons — nothing you type leaves the service (no third-party API), and I could fix the model's judgement directly. That was humbling. Early on it happily gave long, confident, wrong answers high scores and short correct ones low ones (classic length bias). The fix was all in the data, not the model: adding wrong-but-verbose and short-but-correct examples across every score band until length and correctness finally decoupled. It's free to try (no card), with a Pro tier for heavy use. I'd genuinely love your brutal feedback — especially the scoring. Run an interview for a role you know well and tell me: where is it too harsh, or too generous? That's the thing I most want to get right.

Peakblick

Practice tech interviews with an AI that scores you

Practice tech interviews with an AI that scores you