Inspiration

-Preparing for interviews is stressful, especially technical ones. We’ve all had moments where we practiced answers but had no real feedback or realistic environment. We wanted to build something that feels like a real interviewer, not just a question list — a coach that listens, analyzes, and helps you improve through real-time feedback using AI. That’s how Prep67 was born.

What it does

  • Prep67 is a browser-based mock interview coach that uses your webcam and mic to simulate real interviews. It tracks attention and posture using MediaPipe, transcribes your speech, and uses an LLM to evaluate your answers with STAR-method feedback. It also uses an AI interviewer with voice feedback and follow-up questions. Users can skip questions, view summaries, and get clear insights on confidence, focus, and relevance.

How we built it

  • Built with React + Vite, MediaPipe Face Mesh, Web Speech API, and GPT API. We handled video, audio, and AI processing all in-browser with Tailwind UI and a custom session flow that detects when users stop speaking.

Challenges we ran into

  • Detecting when an answer ends, syncing video with speech analysis, and maintaining smooth performance under the hackathon time limits were major challenges.

Accomplishments that we're proud of

  • We built a useful and functional AI interviewer quickly and complete with live feedback and speech, facial tracking, and dynamic question flow.

What we learned

  • Real-time AI interaction requires balancing performance, clarity, and timing. We also learned how to blend multiple AI tools into one seamless experience.

What's next for Prep67

We plan to fully expand Prep67 into a personalized AI interview platform, and: -Add user progress dashboards with detailed analytics and improvement tracking. -Introduce role-specific question sets for software, design, business and many more roles. -Integrate multi-language support for global accessibility. -Offer a university/bootcamp version to help students prepare for technical interviews at scale.

Built With

Share this project:

Updates