NovaPrep

NovaPrep is a conversational AI interview coach that uses Amazon Nova to simulate realistic mock interviews, deliver voice-based follow-ups, and provide actionable feedback for students.

Comment

Inspiration

We built NovaPrep because interview preparation can feel overwhelming, especially for students and early-career job seekers who may not have access to mentors, mock interview partners, or professional coaching. A lot of current tools either feel too static, like reading questions from a list, or too generic, like chatting with a bot that does not really simulate a realistic interview experience.

We wanted to create something more practical and human-centered: an AI interview coach that feels interactive, supportive, and realistic. Our goal was to make interview prep more accessible by combining personalized questions, structured feedback, and voice interaction into one clean experience powered by Amazon Nova.

What it does

NovaPrep is a conversational AI interview coach that helps users practice mock interviews in a more realistic way. Users choose a target role, interview type, and difficulty level, then begin an interview session.

During the session, NovaPrep can:

  • generate tailored interview questions
  • evaluate the user's answers
  • provide a score and detailed feedback
  • suggest a stronger sample answer
  • continue the interview with contextual follow-up questions
  • deliver a final performance summary at the end of the session

We also added voice support so the interviewer can speak responses aloud and the user can answer naturally, making the mock interview feel more like a real conversation instead of a text-only tool.

How we built it

We built the frontend using React and Tailwind CSS to keep the interface modern, fast, and easy to navigate. The backend was built with Node.js and Express, which handled the API routes and all communication with Amazon Bedrock.

The AI side of the project is powered by:

  • Amazon Nova Lite for generating interview questions, evaluating answers, resume-aware personalization, follow-up interviewer responses, and final summaries
  • Amazon Nova Sonic for speech-to-text and spoken interviewer output

We designed the system around structured JSON responses so the frontend could reliably parse and display the AI output. This helped us keep the app simple and stable while still using powerful AI features. We also added optional resume parsing so the interview can be personalized based on the user's background.

Challenges we ran into

One of the biggest challenges was making the interview feel truly conversational. At first, it behaved more like a question-and-answer tool than a realistic interviewer. We had to improve the prompt design and pass conversation history into the model so follow-up questions would feel more natural and connected.

Another major challenge was integrating voice interaction. Working with speech input and spoken interviewer output involved debugging AWS model access, audio handling, streaming issues, and fallback behavior. We had to spend time making sure the voice flow was smooth enough for a live demo while still keeping the app reliable.

We also learned that AI output consistency depends heavily on prompt structure. Getting predictable, well-formatted responses required careful prompt design and strong backend validation.

Accomplishments that we're proud of

We are proud that NovaPrep grew from a simple mock interview idea into a working end-to-end application with a complete user flow. The app does not just generate questions, it creates a full interview experience with setup, conversation, feedback, and summary.

We are especially proud of:

  • building a working MVP that is easy to demo in a few minutes
  • integrating Amazon Nova in a practical way instead of as a simple chatbot wrapper
  • making the interview feel more realistic through contextual follow-up questions
  • adding voice support to create a more immersive experience
  • keeping the project clean, beginner-friendly, and hackathon-ready

What we learned

This project taught us a lot about building real AI products, not just adding AI features. We learned how important it is to focus on reliability, clear UX, and structured outputs when connecting foundation models to a live application.

We also learned how to:

  • work with Amazon Bedrock and Amazon Nova models in a full-stack project
  • design prompts that return predictable JSON for frontend use
  • balance ambitious features like voice and resume-aware personalization with the need for a stable MVP
  • think more carefully about conversational flow, not just single AI responses

Most importantly, we learned that a good AI product is not just about model power. It is about creating an experience that feels useful, understandable, and trustworthy to the user.

What's next for NovaPrep

We see a lot of room to grow NovaPrep beyond the current MVP. Our next steps would be to make the platform even more personalized, interactive, and useful for real job preparation.

Possible next features include:

  • deeper resume-based customization of interview questions
  • stronger voice-first interviewing with fully natural spoken back-and-forth
  • support for more interview formats like system design and product interviews
  • company-specific interview modes
  • progress tracking across multiple sessions
  • role-based preparation plans and improvement recommendations over time

Our long-term vision is for NovaPrep to become an accessible AI coach that helps people practice smarter, build confidence, and walk into real interviews better prepared.

Built With

Share this project:

Updates