InterviewAI

About the project

We've all been there. You spend weeks grinding LeetCode, you ace the technical round — and then you bomb the behavioral interview. Not because you weren't qualified. Because no one ever taught you how to tell your story.

One of our teammates failed the behavioral round at their dream company twice. Same role, same qualifications, different outcome — just because they didn't know how to structure an answer under pressure. That moment stuck with us. The behavioral interview is arguably the highest-leverage 45 minutes in a software engineer's career, yet almost no one practices it seriously. Mock interviews with friends are inconsistent. ChatGPT can't hear you. Expensive coaches charge $300/hour.

We built InterviewAI to be the coaching resource that most people never had access to.

What we learned

We learned that the hardest part of building a voice-first AI product isn't the AI — it's the real-time audio pipeline. Getting ElevenLabs' WebSocket connection to stay stable across browsers, handle mic permissions gracefully, and recover from drops required far more iteration than we expected.

We also learned how much prompt engineering matters. The difference between an AI that gives generic feedback ("good job!") and one that gives specific, actionable coaching ("replace 'we' with 'I' — the interviewer is evaluating your contribution, not the team's") came entirely down to how we structured the system prompt. The researcher on our team spent the first four hours building the knowledge base that made the AI actually useful.

Finally, we learned that a focused 12-hour hackathon forces you to make decisions you'd otherwise agonize over for weeks. Scope cuts that felt painful at hour two looked like the right call by hour ten.

Built With

• amazon-web-services
• css
• elevenlabs
• nanogpt
• nextjs
• python
• react