Intervyou

Inspiration

LeetCode teaches you to solve problems. Being able to solve a leetcode problem and being able to explain the steps and thought behind the code are 2 different skills. Interviewers look not only for the correct and optimal solution but also for the thought process that leads to it. The only way to enhance such a skill is through real mock interviews, which require another person to bounce ideas off of, as well as someone who can provide an unbiased review of what you did well and what needs work. With IntervYou, you can practice these mock interviews on your own, whenever you'd like.

What it does

IntervYou is a voice AI technical interviewer. You pick a topic or paste in any problem, then choose an interviewer's voice and tone. When you're ready, you are able to start your mock AI interview, where the AI will describe the problem to you like a real interview. Furthermore, just like a real interview, you're able to talk to and interact with the AI to clarify questions: edge cases, intended runtime, intended space complexity, etc. You're able to talk through your ideas and get instant feedback on areas for improvement or how the thought process sounds from a third-person perspective. Our AI interviewer will even push back on areas it deems need improvement or could be wrong. Once you feel confident in your interview, you can submit, and you'll be redirected to a summary page. On this page, you'll be given a scorecard with a rating between 0 and 100, and also a hire / no-hire verdict. There are sub-sections within the scorecard where you get a rating of 0 to 5 in certain areas of the interview, like problem solving, communication, correctness, and complexity. Additionally, the review section will also have a section that explains specific areas in which you did well and what areas you could improve upon. The review section will also show your final code submission, a transcript of your interaction with the interviewer, the bugs it found, and an optimal solution to the problem.

How we built it

Next.js, TypeScript, Tailwind and Monaco on the frontend. Gemini Flash in JSON mode with Zod validation runs the interviewer, problem parsing and scoring. ElevenLabs Scribe handles speech-to-text and ElevenLabs TTS gives the interviewer its voice. The code is never executed: Gemini traces it by hand against edge cases, like a human interviewer would.

Challenges we ran into

Designing the system architecture
Figuring out how the frontend, backend, and AI services communicate.
Figuring out when and what to send to AI.
Keeping the scoring honest, since early scorecards were far too generous
Keeping voice latency low and handling Gemini rate limits
We had to define each REST API endpoint’s responsibility, including what data it receives and returns.

Accomplishments that we're proud of

The full voice loop feels like a real conversation. You can paste any question, and it becomes a complete interview in seconds. The scorecard at the end not only grades the code itself on correctness but also grades you on your communication and the overall thought process that led to the final answer.