Brightspace LLM Assistant - Chrome Extension

A powerful Chrome extension that lets you upload course materials, then uses an open-source LLM from Hugging Face to answer questions about deadlines, grading policies, learning objectives, and more.

Features

• 📥 Drag-and-Drop Uploads: Upload PDFs and documents directly from the popup
• 🤖 AI-Powered Answers: Uses an open-source LLM via Hugging Face to answer questions about your course materials
• 💬 Interactive Chat: Conversational interface to ask questions about:
  • Assignment and exam deadlines
  • Grading policies and rubrics
  • Course learning objectives
  • Course requirements and policies
  • And more!
• 💾 Drive-Backed Storage: Uploads files to Google Drive for future RAG
• 🧠 Optional Vector API: Store embeddings in MongoDB Atlas and query top-K chunks
• 🔒 Private: Processes files locally; only sends text summaries to the LLM API

Prerequisites

1. Chrome Browser (version 88 or later)
2. Hugging Face API Key (free tier available)
   • Sign up at huggingface.co
   • Get your API key from huggingface.co/settings/tokens
   • Accept the terms for the selected model access
3. Google Drive OAuth Client ID
   • Add your client ID to manifest.json (oauth2.client_id)
   • Scope used: https://www.googleapis.com/auth/drive.file
4. (Optional) Vector API Server
   • Node.js 18+ recommended for the MongoDB driver

Installation

Step 1: Download/Clone the Extension

cd /Users/parkhiagarwal/Downloads/LMS

Step 2: Load the Extension in Chrome

1. Open Chrome and go to chrome://extensions/
2. Enable Developer mode (toggle in top right)
3. Click "Load unpacked"
4. Select the LMS extension folder
5. The extension should appear in your Chrome extensions list

Step 3: Configure the Extension

1. Click the extension icon in the Chrome toolbar
2. Enter your Hugging Face API key in the "Hugging Face API Key" field
3. Click Save

Usage

Uploading Files

1. Click the extension popup icon
2. Click Authorize Drive when prompted
3. Drag and drop files into the upload area
4. Files are uploaded to Google Drive immediately

Asking Questions

1. Once files are uploaded, type your question in the text area
2. Click "Ask Question" or press Shift+Enter
3. The extension will:
   • Extract text from the uploaded files
   • Send it to the LLM with your question
   • Display the answer in the chat

Quick Question Templates

Use the suggested questions to quickly ask common queries:

• 📅 Upcoming deadlines - Get all assignment and exam dates
• 📊 Grading policy - Learn how your grade is calculated
• 🎯 Learning objectives - Understand course goals

File Structure

LMS/
├── manifest.json          # Extension configuration
├── background.js          # Service worker (LLM communication)
├── popup.html             # Extension UI
├── popup.js               # Popup logic
├── popup.css              # Styling
├── utils.js               # Text extraction utilities
├── README.md              # This file
└── icons/                 # Extension icons
    ├── icon16.png
    ├── icon48.png
    └── icon128.png

How It Works

1. Background Service Worker (background.js):

   • Receives questions from the popup
   • Fetches and extracts text from uploaded files
   • Creates a context-aware prompt
   • Calls Hugging Face API with the selected model
   • Returns answers to the popup

2. Popup UI (popup.html, popup.js, popup.css):

   • User interface for interacting with the extension
   • Uploads files to Google Drive
   • Displays uploaded files
   • Chat interface for questions and answers
   • Stores API key securely in Chrome storage

LLM Configuration

Default Model

The extension uses a small language model (SLM) by default via Hugging Face Inference API:

• google/gemma-2-2b-it

Other Available Models

You can modify background.js to use alternative models:

• meta-llama/Llama-3.1-8B-Instruct (balanced speed/quality)
• mistralai/Mistral-7B-Instruct-v0.3 (strong general-purpose)
• Other open-source text-only models available on Hugging Face

RAG & Embeddings

The extension uses a lightweight RAG pipeline to ground answers in your course files:

• Embeddings model: Qwen/Qwen3-Embedding-8B via Hugging Face Router
• Chunking: 1200 characters with 150 overlap
• Retrieval: top 8 chunks per query
• Vector store: saved to vector-store.json in your Google Drive folder Brightspace LLM Assistant

Optional Vector API (MongoDB Atlas)

You can store embeddings in MongoDB Atlas and retrieve top-K chunks via a local API server.

1. Set up a vector index in MongoDB Atlas on the embedding field.
2. Configure the server:
   • Copy server/.env.example to server/.env
   • Set MONGODB_URI, MONGODB_DB, MONGODB_COLLECTION, and VECTOR_INDEX_NAME
3. Start the server:
   • cd server && npm install
   • npm start
4. In the extension popup, set Vector API URL (e.g., http://localhost:3000/api).
5. If you set API_KEY in server/.env, set the same value in Vector API key in the popup.

When the Vector API URL is set, the extension stores embeddings in MongoDB and queries top-K chunks from the API.

Local Alternative (Ollama)

For completely private inference without API keys:

1. Install Ollama
2. Run ollama pull llama3.2:3b-instruct
3. Start Ollama: ollama serve
4. Enable local mode in the popup and set the model name

Troubleshooting

API Key errors

• Verify your Hugging Face API key is correct
• Make sure you have accepted the selected model terms on Hugging Face
• Check that your free tier account has API access active

No response from LLM

• Check your internet connection
• Verify the file wasn't too large (token limits apply)
• Try a simpler question first
• Check Hugging Face service status

Drive upload issues

• Make sure you authorized Google Drive in the popup
• Check browser console (F12) for OAuth errors
• Verify your OAuth client ID in manifest.json

Vector API errors

• Invalid API key: Set Vector API key in the popup to match API_KEY in server/.env, or remove API_KEY to disable auth.
• Failed to connect to MongoDB: Double-check MONGODB_URI credentials and Atlas IP allowlist.

Privacy & Security

• ✅ Files are processed locally for text extraction
• ✅ Uploaded files are stored in your Google Drive folder Brightspace LLM Assistant
• ✅ Vector index is stored in vector-store.json in the same Drive folder (if Vector API URL is not set)
• ✅ Vector API mode stores embeddings in MongoDB Atlas
• ✅ API key is stored only in Chrome local storage
• ✅ Only text content is sent to the LLM API
• ⚠️ Your Hugging Face API key is visible to Hugging Face services
• ⚠️ Don't share your API key with others

Limitations

• Token Limits: LLM has context window limits (~2000 tokens for free tier)
• File Types: Currently best with text-based files (PDF, TXT, DOCX); requires pdf.js for PDF extraction
• Accuracy: LLM responses are AI-generated and may contain errors
• Rate Limits: Free tier Hugging Face has rate limits
• Large Files: Very large files may be truncated to fit token limits

Future Enhancements

• Add support for multiple file formats (images, audio)
• Implement local PDF.js for better PDF extraction
• Add mammoth.js for better DOCX support
• Support for document summarization
• Custom prompt templates for different question types
• Integration with other learning platforms (Canvas, Moodle, etc.)
• Multi-language support
• Offline mode with local LLM

Contributing

Feel free to extend and modify this extension:

1. Add better file parsing (pdf.js, mammoth.js)
2. Support more file types
3. Improve the UI/UX
4. Add more LLM providers
5. Optimize token usage

License

This project is provided as-is for educational purposes.

Support & Questions

For issues or questions:

1. Check the troubleshooting section
2. Review browser console errors (F12)
3. Check Hugging Face API documentation
4. Review Chrome extension documentation

Resources

• Chrome Extension Documentation
• Hugging Face API Docs
• Gemma Model Card
• Google Drive API

Disclaimer

This extension is not affiliated with Brightspace, Meta, or Hugging Face. Use it responsibly and in accordance with your institution's policies.

brightspace_helper