Prepwiser

Prepwiser App Logo

Inspiration

We wanted to recreate that feeling of practicing interviews with your homies—where you hype each other up, ask challenging questions, and build real confidence before the big day. Interview prep often feels robotic, isolating, and stressful. With Prepwiser, our goal was to make interview practice feel personal, engaging, and empowering—just like a friend helping you prepare, but powered by AI.

What it does

Prepwiser is an AI-powered mock interview platform that helps users practice behavioral and technical interviews in a personalized, conversational, and engaging way.

Key features include:

🎥 AI Interviewers with Tavus Avatars: Lifelike video-based AI interviewers who ask questions and interact with users in real-time.
🏢 Fully AI-Generated, Personalized Questions: The AI dynamically generates interview questions based on user-provided inputs like Role, Company, Resume, Job Description, Interview Type, Interview Duration, and Additional Notes.
💬 Real-Time Voice Conversations: Users can speak their answers, and the AI listens, analyzes, and responds—creating a natural interview flow.
📝 Instant AI Feedback: After the session, users receive AI-generated feedback and suggestions for improvement.
🔗 External Coding Practice Links: For technical interview prep, users get links to platforms like LeetCode, CodeSignal, etc., to practice coding questions until we build our own integrated coding module.

How we built it

Frontend: React (TypeScript) with Tailwind CSS for a responsive, clean user interface.
Backend: FastAPI for API management, session control, and user data handling.
AI & Voice Pipeline:
- LiveKit AI Agents: Powering real-time voice interaction, speech-to-text (STT), and LLM-driven dynamic question and feedback generation.
- Tavus AI Avatars: Delivering realistic, video-based AI interviewer responses with lifelike text-to-speech (TTS).
- LLMs: Used for personalized question generation, adaptive dialogue management, and behavioral feedback.
Database: Supabase for user authentication, interview history, and performance tracking.

Challenges we ran into

Building low-latency, real-time voice conversations between users and AI agents.
Making AI-generated questions truly personalized and role/company-specific using diverse input fields.
Seamlessly syncing LiveKit voice streams with Tavus video avatars for a smooth conversational flow.
Managing multi-step session states across behavioral interviews, user responses, feedback analysis, and coding resource recommendations.

Accomplishments that we're proud of

Successfully launched video-based AI interviewers using Tavus avatars.
Created a fully AI-driven question generation pipeline with zero reliance on static question banks.
Built an end-to-end interview flow that covers user input collection, real-time AI interviewing, voice transcription, LLM-based analysis, and personalized feedback.
Provided external coding resources while maintaining a seamless user experience for both behavioral and technical interview prep.

What we learned

How to orchestrate complex AI pipelines that combine real-time voice, LLM reasoning, TTS video delivery, and user input-driven personalization.
Prompt engineering techniques for generating context-aware and highly relevant interview questions.
Building state management for multi-step, multi-modal interview sessions.
User experience design for keeping candidates engaged and making interview prep feel less stressful and more conversational.

What's next for Prepwiser

Building and integrating our own coding assessment platform so users can complete technical challenges directly inside Prepwiser.
Adding more customization options for AI interviewers, like tone adjustment (friendly, serious, etc.).
Expanding feedback analysis to provide quantitative scoring, improvement areas, and skill gap reports.
Implementing gamified progress tracking (badges, streaks, XP) to motivate regular practice.
Launching a mobile app version for on-the-go interview prep.