📐 Math Mentor

An AI-powered JEE mathematics tutor with multimodal input, RAG-backed knowledge retrieval, and a multi-agent reasoning pipeline.

---

✨ Features

• Multimodal Input — type a problem, upload an image, or speak into the mic
• 5-Agent Pipeline — each agent has a single responsibility for clean, traceable reasoning
• RAG Knowledge Base — curated JEE formulas, worked examples, and common mistakes
• Memory Layer — learns from past interactions and surfaces similar solved problems
• Human-in-the-Loop — flags low-confidence solutions for manual review before delivery
• Dark UI — clean Streamlit interface with agent trace, confidence badge, and feedback system

---

🧠 Architecture

Input (text / image / audio)
        ↓
  Input Processor        ← OCR or Whisper transcription
        ↓
     Parser              ← extracts problem, topic, variables
        ↓
     Router              ← classifies: algebra / calculus / probability / linear algebra
        ↓
     Solver              ← RAG retrieval + memory hint + LLaMA 3.3 70B
        ↓
    Verifier             ← confidence scoring, HITL trigger if low confidence
        ↓
    Explainer            ← step-by-step student-friendly explanation

---

🗂️ Project Structure

math-mentor/
├── agents/
│   ├── parser.py         # Extracts structured problem from raw input
│   ├── router.py         # Classifies topic
│   ├── solver.py         # RAG + memory backed solver
│   ├── verifier.py       # Confidence scoring + HITL
│   └── explainer.py      # Student-facing explanation
├── rag/
│   └── retriever.py      # ChromaDB vector store builder + retriever
├── memory/
│   └── store.py          # SQLite-backed interaction memory
├── tools/
│   ├── image_parser.py   # Groq vision OCR
│   ├── audio_parser.py   # Groq Whisper transcription
│   └── embeddings.py     # Embedding model wrapper
├── kb/                   # Markdown knowledge base (JEE formulas + examples)
├── graph.py              # LangGraph state machine
├── app.py                # Streamlit UI
└── .env.example

---

⚙️ Setup

1. Clone the repo

git clone https://github.com/Noob-Master-22/math-mentor.git
cd math-mentor

2. Create virtual environment

python3 -m venv venv
source venv/bin/activate

3. Install dependencies

pip install -r requirements.txt

4. Add your API keys

cp .env.example .env

Edit .env:

GROQ_API_KEY=your_groq_api_key_here

5. Build the vector index

python3 rag/retriever.py

6. Run the app

streamlit run app.py

---

🧪 Input Modes

Mode	How
✍️ Text	Type any JEE math problem directly
🖼️ Image	Upload a photo or screenshot of a handwritten/printed problem
🎙️ Audio	Record live via mic or upload an .m4a file

---

🛠️ Tech Stack

Layer	Technology
LLM	Groq — LLaMA 3.3 70B Versatile
Speech-to-Text	Groq — Whisper Large v3
Vision / OCR	Groq — LLaMA Vision
Orchestration	LangGraph
Vector Store	ChromaDB
Embeddings	HuggingFace / LangChain
Memory	SQLite
UI	Streamlit

---

📄 License

Copyright (c) 2026 Satvik Reddy. All rights reserved.

This project and its source code are made available for evaluation and review purposes only.

• You may view and run this code locally for assessment
• You may not copy, modify, distribute, sublicense, or use any part of this code in your own projects
• You may not use this code commercially or claim it as your own work

Unauthorized use, reproduction, or distribution of this code is strictly prohibited.