# VivaMate — An AI Viva Partner Built for a Friend

VivaMate is an AI-powered platform designed to assist students preparing for technical university vivas. The challenge the creator faced was that while studying from notes is one thing, actually answering questions and getting meaningful feedback is another. Finding someone available to conduct a mock viva was often impractical. To address this, VivaMate was built as an on-demand viva practice partner.

Key features of VivaMate include:

1. Uploading study material as a PDF or TXT file

2. Generating viva questions based on the uploaded material

3. Selecting the number and difficulty of questions

4. Taking a one-question-at-a-time mock viva

5. Submitting answers for AI evaluation

6. Receiving a score, feedback, ideal answer, and follow-up question

7. Viewing a final performance report showing strengths and areas to improve

The application is deployed using a React/Vite frontend on Vercel and a FastAPI backend on Render. The core model used for AI functionality is Qwen3 4B, an open-weight model. Qwen3 4B is used for two main purposes: generating viva questions and evaluating student answers. During local development, Qwen3 4B is run locally through Ollama, allowing for direct development and testing of the AI workflow on the creator's machine.

For deployment, the same AI service abstraction enables the use of Hugging Face Inference Providers with Qwen3 4B.

VivaMate's architecture separates the AI integration from the rest of the application, allowing for flexibility in changing the model or inference provider without rewriting the entire system. The main technologies used in the application are React + Vite, FastAPI, Python, Ollama, Qwen3 4B, Hugging Face Inference Providers, Pydantic, PyPDF, HTML/CSS/JavaScript.

The AI is utilized for two critical aspects of the application: generating viva questions and evaluating student answers. VivaMate follows a simple practice loop where students upload material, generate questions, answer questions one at a time, receive evaluation, and improve based on the feedback provided. The application's design ensures that students can practice repeatedly before the actual viva, without relying on a professor or a real viva.