StuBud: Turn Your Study Material Into an Active Learning Workspace with Gemma(LOCAL LEARNING LAB)

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

I built StuBud, an AI-powered learning workspace designed around a problem I noticed in my own friend group: my friend was struggling to score well in exams even though he had the study material.

The problem wasn't simply access to notes. Before an exam, there was too much material to go through and not enough time to turn it into something useful for revision.

So I built StuBud around the idea of a one-shot revision workspace.

A student can upload their study material and turn the same sources into different ways of learning:

Flashcards for quick active recall

Quiz for testing understanding

Mind Map for seeing how concepts connect

Study Guide for structured revision

Exam Prep for focusing on topics that need more attention

Audiobook for listening-based revision

The goal is not to replace studying with AI.

The goal is to take the material a student already needs to study and quickly turn it into a set of revision tools they can actually use before an exam.

The idea behind StuBud is simple:

Just Upload your notes once.Turn them into multiple ways to revise.

I built it for students like my friend who need a practical way to go from a pile of notes to a focused revision session, especially when an exam is approaching.

Live Demo: https://stubud-c0ps.onrender.com/

Try now!!!!

GitHub Repository: https://github.com/kachamsiddarth/StuBud

StuBud is built around open-weight AI, with Gemma 4 at the core of the learning experience.

My development process went through several Processes because getting a model to generate text is very different from getting it to reliably generate output that a real application can use.

I initially experimented with Transformers.js and a local ONNX version of Gemma. The goal was to keep inference inside the JavaScript ecosystem and run the model locally.

The model could run, but I ran into practical problems when connecting it to the actual StuBud workflows. Some responses did not consistently match the structured formats required by features such as Flashcards, Quiz, and Study Guide, and CPU inference could also take a long time.

Instead of removing local AI, I switched the local inference runtime to Ollama and used the quantized Gemma 4 E2B model.

This gave me a much more practical local Gemma setup, although CPU inference was still slow for larger generations. That led me to add provider fallbacks, timeouts, structured-output validation, and repair logic instead of allowing individual features to depend directly on one model call.

StuBud uses a shared AI provider layer rather than having every feature directly call a model.

StuBud Feature
      ↓
AI Service / Provider Router
      ↓
┌───────────────┬───────────────┐
│               │               │
Gemma 4     Open Model/Groq   Local Gemma 4
                                ↓
                         Local Inference

I enjoy making my projects open source because I believe useful technology should be accessible to anyone, regardless of where they live.

I have seen many study tools that are paid or focus on individual features, but I wanted StuBud to focus on a different problem: helping a student turn their own notes into a focused one-hour revision session before an exam.

Using open-weight AI like Gemma 4 gives me the freedom to experiment with local inference, different providers, and different ways of building the learning experience without being locked into one closed AI system.

Making StuBud open source also means that someone from any corner of the world can access the project, learn from it, improve it, or build something new on top of it.

I used GitHub Copilot along with development plugins and skills during the build for implementation, debugging, frontend work, and a lot of other stuff.

For verification, I used Node.js's built-in test runner with tsx, TypeScript type checking, builds, and real AI workflow testing across the major StuBud features.

So , This helped me develop the project faster and made it much easier to identify and fix bugs during development.

Best Use of Gemma — $200 — AI Generation

Best Use of Render — $200 — Deployment

Best Use of MongoDB Atlas — $100 — Database

Best Use of ElevenLabs — $100 — Text-to-Speech

Best Use of GitHub Copilot — $100 — Development