DocMind 🧠

Chat with any PDF using AI — powered by MistralAI, LangChain, and ChromaDB.

---

What is DocMind?

DocMind is a Retrieval-Augmented Generation (RAG) application that lets you upload any PDF and have a natural conversation with it. Ask questions, extract insights, or summarize content — the AI answers using only what's in your document.

No hallucinations. No guessing. Just your document, intelligently searched.

---

How It Works

┌─────────────────────────────────────────────────────────────────┐
│                        DocMind Flow                             │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│   📄 PDF Upload                                                 │
│        │                                                        │
│        ▼                                                        │
│   PyPDFLoader → loads all pages as documents                    │
│        │                                                        │
│        ▼                                                        │
│   RecursiveCharacterTextSplitter                                │
│   (chunk_size=1000, overlap=200)                                │
│        │                                                        │
│        ▼                                                        │
│   MistralAI Embeddings (mistral-embed)                          │
│   → converts each chunk into a vector                           │
│        │                                                        │
│        ▼                                                        │
│   ChromaDB → stores all vectors locally                         │
│                                                                 │
│   ─────────────────────────────────────────                     │
│                                                                 │
│   💬 User Question                                              │
│        │                                                        │
│        ▼                                                        │
│   MMR Retriever (k=4, fetch_k=10, lambda=0.5)                  │
│   → finds the 4 most relevant & diverse chunks                  │
│        │                                                        │
│        ▼                                                        │
│   ChatPromptTemplate                                            │
│   → injects context + question into prompt                      │
│        │                                                        │
│        ▼                                                        │
│   MistralAI LLM (mistral-small-2506)                            │
│   → generates answer grounded in context                        │
│        │                                                        │
│        ▼                                                        │
│   ✅ Answer displayed in chat UI                                │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘

---

Tech Stack

Layer	Technology	Purpose
Frontend	HTML, CSS, Vanilla JS	Chat UI with PDF upload
Backend	Flask (Python)	REST API server
LLM	MistralAI mistral-small-2506	Answer generation
Embeddings	MistralAI mistral-embed	Text → vector conversion
Vector Store	ChromaDB	Storing & searching embeddings
RAG Framework	LangChain	Orchestrating the entire pipeline
PDF Loader	PyPDFLoader	Parsing PDF pages
Text Splitter	RecursiveCharacterTextSplitter	Chunking documents
Retrieval	MMR (Maximal Marginal Relevance)	Diverse, relevant chunk retrieval

---

Project Structure

DocMind/
├── app.py                  # Flask backend — API routes
├── index.html              # Frontend UI
├── create_database.py      # Standalone script to pre-index a PDF
├── main.py                 # CLI version of the RAG chat
├── requirements.txt        # Python dependencies
├── .env                    # API keys (never committed)
├── .gitignore
└── chroma-db/              # Local vector store (auto-generated)

---

Local Setup

1. Clone the repository

git clone https://github.com/YOUR_USERNAME/docmind.git
cd docmind

2. Create and activate a virtual environment

python -m venv venv
source venv/bin/activate        # Mac/Linux
venv\Scripts\activate           # Windows

3. Install dependencies

pip install -r requirements.txt

4. Set up environment variables

Create a .env file in the root folder:

MISTRAL_API_KEY=your_mistral_api_key_here

Get your free API key at console.mistral.ai

5. Run the app

python app.py

Open your browser at http://localhost:5010

---

Usage

1. Open the app in your browser
2. Click "Click or drag a PDF here" in the sidebar
3. Wait for the document to be indexed (you'll see a confirmation message)
4. Type your question in the chat input and press Enter
5. DocMind retrieves the most relevant chunks and answers using only your document

---

API Endpoints

Method	Endpoint	Description
GET	/	Serves the frontend UI
POST	/upload	Upload and index a PDF file
POST	/chat	Send a question, get an answer
GET	/status	Check if a document is loaded

---

Environment Variables

Variable	Required	Description
MISTRAL_API_KEY	✅ Yes	Your MistralAI API key
PORT	Auto (Railway)	Port override for deployment

---

Live Demo

🚀 https://docmind-etyl.onrender.com/

---

Author

Built by Henil Bhavsar

---

License

MIT License — free to use, modify, and distribute.