PDF ChatBot - Chat with Any PDF using AI

An AI-powered PDF ChatBot with a dark chat UI. Upload any PDF, ask questions in a conversational interface, and get grounded answers powered by Llama 3.3 70B via OpenRouter.

Built as a portfolio project - fast, lightweight, zero local GPU required.

---

Features

• Upload & parse any PDF with PyMuPDF
• Smart text chunking via LangChain
• Semantic embeddings using all-MiniLM-L6-v2
• Vector storage & retrieval with Pinecone (serverless)
• Answers from Llama 3.3 70B via OpenRouter (free tier)
• Persistent chat history with full conversation UI
• Dark GitHub-inspired theme
• Auto-retry on rate limits, batch upsert for large PDFs

---

Architecture

PDF Upload
    |
    v
PyMuPDF -- extract raw text
    |
    v
LangChain RecursiveCharacterTextSplitter -- chunk (500 tokens, 50 overlap)
    |
    v
SentenceTransformers all-MiniLM-L6-v2 -- embed chunks -> 384-dim vectors
    |
    v
Pinecone Serverless -- store & query vectors (top-5 retrieval)
    |
    v
OpenRouter -> Llama 3.3 70B Instruct -- generate grounded answer
    |
    v
Streamlit Dark Chat UI -- display in conversation thread

---

Tech Stack

Layer	Technology
Frontend	Streamlit with custom dark CSS
PDF Parsing	PyMuPDF
Text Splitting	LangChain RecursiveCharacterTextSplitter
Embeddings	SentenceTransformers all-MiniLM-L6-v2
Vector DB	Pinecone Serverless (free tier)
LLM	Llama 3.3 70B via OpenRouter (free)
API Client	OpenAI Python SDK

---

Project Structure

app/
├── app.py                  # Streamlit UI entry point
├── config.py               # All constants (model names, chunk sizes, etc.)
├── pdf_utils.py            # PDF extraction and text chunking
├── embedder.py             # SentenceTransformer loading and vector encoding
├── vector_store.py         # Pinecone connect, upsert, and query
├── llm.py                  # OpenRouter API call with retry logic
├── .streamlit/
│   └── config.toml         # Dark theme + server config for Streamlit Cloud
├── .env.example            # Template -- copy to .env and fill in keys
├── .gitignore              # Excludes .env, .venv, temp.pdf, __pycache__
├── requirements.txt        # Minimal direct dependencies (7 packages)
└── README.md               # This file

---

Local Setup

1. Clone the repo

git clone https://github.com/RohitDSonawane/PDFChatBot.git
cd PDFChatBot

2. Create and activate a virtual environment

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

3. Install dependencies

pip install -r requirements.txt

4. Set up environment variables

cp .env.example .env

Then edit .env and add your keys:

PINECONE_API_KEY=your-pinecone-api-key
OPENROUTER_API_KEY=your-openrouter-api-key

• Get Pinecone key: app.pinecone.io -> API Keys
• Get OpenRouter key (free): openrouter.ai -> Keys

5. Run the app

streamlit run app.py

---

Deploy on Streamlit Community Cloud (Free)

1. Push this repo to GitHub (public)
2. Go to share.streamlit.io -> sign in with GitHub
3. Click New app -> select this repo -> set app.py as entry point
4. Go to Advanced settings -> Secrets and add:

PINECONE_API_KEY = "your-pinecone-api-key"
OPENROUTER_API_KEY = "your-openrouter-api-key"

5. Click Deploy -- live in ~2 minutes

---

Author

Rohit Sonawane

• LinkedIn
• GitHub

---

Star this repo if you found it useful! Feedback, issues, and PRs are welcome.