Intervue

Inspiration

Most interview prep tools are static: you read a question on a screen and type an answer. But real interviews are dynamic, verbal, and unpredictable. Candidates often struggle not with what they know, but with how they communicate it under pressure.

An AI interviewer asks questions and follows up in real time
Users explain their thinking out loud while solving problems
The system adapts dynamically based on responses, just like a real interviewer
After the session, users receive feedback on:
- communication clarity
- structure of answers
- technical reasoning
- overall performance

What it does

Intervue is a live mock interview platform that simulates real technical interviews through voice interaction. It adapts questions based on your responses and gives feedback on your communication, structure, and overall performance.

How we built it

Frontend: React, TypeScript, Tailwind CSS, Clerk
Backend: FastAPI, Python, WebSocket, Judge0 (code execution)
Voice/AI: ElevenLabs, Tavily, Featherless (Gemini 2.0 Flash), Groq
Database: MongoDB

Challenges we ran into

Real-time adaptation: Handling user speech, processing transcripts, and deciding the next action (follow-up, pushback, next question) required careful orchestration.
Voice interaction reliability: Ensuring smooth back-and-forth conversation with minimal latency was challenging, especially when coordinating between multiple services.

Accomplishments that we're proud of

Built a fully functional live mock interview system end-to-end
Created an experience that feels interactive and realistic, not scripted
Integrated voice, coding, and feedback into a single seamless flow

What we learned

We learned how to design systems that adapt to users in real time and how important communication is in technical interviews.
The importance of prompt design for your LLM Model to give the right outputs for your specific app.

What's next for Intervue

We plan to improve feedback accuracy, add more question types, and track user progress over time. We also want to explore adding incentives, like streaks, to keep users engaged.