Ai9campus/edubot-textbook

EduBot — AI Document Intelligence

Multilingual Retrieval-Augmented Generation (RAG) Demo

Ask any question from your textbook in English, Hindi, Telugu, or Urdu. Every answer is grounded in the actual textbook with page citations.

Technology Stack

• LLM: Groq LLaMA-3.3-70B (free, ultra-fast)
• Embeddings: paraphrase-multilingual-MiniLM-L12-v2 (50+ languages)
• Vector Store: ChromaDB (persistent cosine similarity search)
• PDF Extraction: PyMuPDF (best math symbol support)
• UI: Gradio on HuggingFace Spaces

Setup Instructions

Step 1 — Add your PDF

Upload your textbook PDF and rename it to textbook.pdf. Place it inside a folder called data/ in the Space files. Final path should be: data/textbook.pdf

Step 2 — Add GROQ_API_KEY secret

Go to Space Settings → Variables and secrets → Add GROQ_API_KEY Get your free key from: https://console.groq.com

Step 3 — The app builds automatically

On first startup, EduBot reads your PDF, chunks it, embeds all chunks, and saves the index to ChromaDB. This takes 5-10 minutes once. After that, every restart loads the cached index instantly.

Languages Supported

Language	Script	Generation Method
English	Latin	LLaMA-3.3-70B direct
Hindi	Devanagari	LLaMA-3.3-70B direct
Telugu	Telugu script	LLaMA-3.3-70B direct
Urdu	Arabic script	LLaMA-3.3-70B direct

How It Works

1. Question is embedded using multilingual model
2. ChromaDB retrieves top-4 most relevant textbook passages
3. Passages + question sent to LLaMA-3.3-70B via Groq API
4. Model answers using ONLY the textbook — no hallucination
5. Page numbers shown as citations

Cost

• HuggingFace Spaces: Free (CPU tier)
• Groq API: Free (within generous rate limits)
• Total cost: $0