NovaMentor - AI Interview Coach

NovaMentor

Inspiration

Preparing for technical interviews is challenging for many students and job seekers. Most preparation methods rely on static question banks or watching interview videos, which do not provide real-time feedback or simulate the experience of a real interview.

I wanted to build an AI system that acts like a real interviewer — asking personalized questions, evaluating responses, and providing actionable feedback. With the reasoning capabilities of Amazon Nova, it became possible to create an intelligent interview coach that adapts to each candidate’s background and skills.

NovaMentor was built to make interview preparation more interactive, personalized, and accessible.

What it does

NovaMentor is an AI-powered mock interview coach built using Amazon Nova Lite via AWS Bedrock.

The platform simulates real technical interviews by:

Generating personalized interview questions based on the candidate’s profile
Evaluating answers using AI reasoning
Providing structured feedback and scoring
Suggesting improvements and follow-up questions
Generating a final interview session summary

The system supports multiple interview modes:

Interest-based interview questions
Scenario-based problem solving
System design discussions
DSA / LeetCode-style coding questions with a built-in code editor

This creates a realistic interview practice environment powered by AI.

How we built it

NovaMentor uses a modular architecture combining a Streamlit frontend with Amazon Nova Lite via AWS Bedrock.

Frontend

Streamlit UI with modern dark theme
Streamlit-Ace code editor for coding questions

Backend

Python application logic
Prompt engineering for question generation and evaluation
Candidate context system to personalize interviews

AI Layer

Amazon Nova Lite through AWS Bedrock
Used for generating interview questions, evaluating answers, and producing interview summaries

The workflow:

Candidate Input → Streamlit UI → Python Backend → AWS Bedrock → Amazon Nova Lite → AI Feedback

A flexible LLM abstraction layer was implemented so the system can switch between Amazon Nova (production) and Ollama (local development).

Challenges we ran into

One of the biggest challenges was designing prompts that evaluate answers fairly and consistently. AI models can sometimes be overly generous, so we implemented strict scoring guidelines to ensure realistic feedback.

Another challenge was maintaining interview context so that the AI could generate relevant questions based on the candidate’s profile and previous responses.

We also worked on integrating a coding editor for DSA-style questions while keeping the user experience simple and interactive.