This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend
What I Built
VivaBuddy is an AI examiner that runs fully on my laptop. I built it for my friend [Name], who has a [subject] viva on [date]. [Name] prepares by re-reading notes, but freezes when the examiner asks a follow-up question. VivaBuddy asks questions from [Name]'s own notes and adapts: a strong answer gets a harder follow-up, a weak one gets an easier question from another angle.
Demo
https://drive.google.com/file/d/1w60YCTo1ogxUJZ3-YGB6uqfGfTmxyDPL/view?usp=sharing
Code
GITHUB REPO LINK :- https://github.com/Jinal301409/Hacktoberfest-vivabuddy.git
How I Built It
Notes are split into sections and embedded locally with nomic-embed-text.
- Gemma (
gemma3:4b) runs through Ollama and grades each answer only against those notes. - Scores are averaged per topic in code, and the model only writes the final recommendation.
- Stack: React, Node/Express, MongoDB Atlas.
Why Does Open Innovation Matter?
The model runs locally, so [Name]'s notes are never sent to an AI company.
- It costs nothing per question.
- I could swap models: I started on a 1B Gemma, then moved to 4B when the small one graded too generously.
- Honest note: MongoDB Atlas is cloud-hosted, so notes and scores are stored there. Only the AI model runs locally.
Top comments (0)