PDF Chat Assistant

A production-grade RAG (Retrieval-Augmented Generation) application that allows you to have intelligent, grounded conversations with your PDF documents. Built with LangChain, ChromaDB, and Claude API (Anthropic).

---

✨ Features

• Local Embedding Processing: Uses all-MiniLM-L6-v2 to process vectors locally on your CPU—no data leaves your machine until the final LLM prompt.
• State-of-the-Art Reasoning: Powered by Claude 4.7, providing high-accuracy answers based strictly on your context.
• Persistent Storage: ChromaDB integration ensures that once a PDF is indexed, it stays indexed across sessions.
• Source Attribution: Every answer includes a "View Sources" dropdown, showing you the exact page numbers and text snippets used to generate the response.
• Dockerized: Fully containerized for easy deployment and consistent environments.

---

📸 Demo

🛠️ Project Structure

.
├── src/
│   ├── ingestion/      # PDF Loading and Text Splitting
│   ├── retrieval/      # Embeddings and Vector Store logic
│   ├── generation/     # LLM factory and Prompt templates
│   └── services/       # RAG Pipeline orchestration
|
├── tests/              # Pytest suite
├── app.py              # Streamlit Entry Point
├── Dockerfile          # Container configuration
└── .env.example        # Template for secrets

---

🏗️ Architecture

The application is split into two distinct phases — Ingestion (run once per PDF) and Querying (run on every user question).

╔══════════════════════════════ INGESTION PIPELINE ═══════════════════════════╗
║                                                                             ║
║   📄 PDF File                                                               ║
║       │                                                                     ║
║       ▼                                                                     ║
║   [ PdfReader  ]        pypdf.PdfReader extracts raw text page by page      ║
║       │                                                                     ║
║       ▼                                                                     ║
║   [ Text Chunker ]      RecursiveCharacterTextSplitter slices text into     ║
║       │                 overlapping chunks (CHUNK_SIZE / CHUNK_OVERLAP)     ║
║       ▼                                                                     ║
║   [ Embedder ]          all-MiniLM-L6-v2 converts each chunk into a         ║
║       │                 384-dim vector — runs locally on CPU                ║
║       ▼                                                                     ║
║   [ ChromaDB ]          Vectors + metadata stored persistently on disk      ║
║                                                                             ║
╚═════════════════════════════════════════════════════════════════════════════╝
 
╔══════════════════════════════ QUERY PIPELINE ═══════════════════════════════╗
║                                                                             ║
║   💬 User Question                                                          ║
║       │                                                                     ║
║       ▼                                                                     ║
║   [ Embedder ]          Same model encodes the question into a vector       ║
║       │                                                                     ║
║       ▼                                                                     ║
║   [ ChromaDB Retriever ] Top-K most similar chunks fetched by cosine        ║
║       │                  similarity (default: TOP_K_RESULTS=4)              ║
║       ▼                                                                     ║
║   [ Prompt Builder ]    Question + retrieved chunks assembled into a        ║
║       │                 grounded prompt with strict context instructions    ║
║       ▼                                                                     ║
║   [ Claude 4.7 ]        LLM generates an answer — strictly from context     ║
║       │                                                                     ║
║       ▼                                                                     ║
║   💡 Answer + Sources   Response shown with page-level source attribution   ║
║                                                                             ║
╚═════════════════════════════════════════════════════════════════════════════╝

Why two phases? Ingestion is expensive (I/O + embedding) but only runs once. Querying is fast because retrieval and generation are the only steps at runtime.

---

🚀 Quick Start

1. Prerequisites

• Python 3.12+
• An Anthropic API Key

2. Installation

# Clone the repository
git clone https://github.com/aysa-dev1/chat_with_your_documents.git
cd chat_with_your_documents

# Create a virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

3. Environment Setup

Copy .env.example to .env in the root directory and fill in your values:

ANTHROPIC_API_KEY=your_sk_key_here
ANTHROPIC_MODEL=claude-opus-4-7
MAX_TOKENS=1024
CHUNK_SIZE=1000
CHUNK_OVERLAP=200
CHROMA_PERSIST_DIR=./data/chroma
TOP_K_RESULTS=4

4. Run the App

streamlit run app.py

---

🐳 Docker Deployment

If you prefer to run with Docker:

# Build the image
docker build -t chat-with-documents .

# Run the container
docker run -p 8501:8501 --env-file .env chat-with-documents

---

🧪 Testing

The project includes a robust test suite covering ingestion logic, prompt construction, and the RAG pipeline orchestration using pytest and pytest-mock.

pytest tests/

---

🛡️ Key Guardrails

• No Hallucinations: The system message explicitly instructs the LLM to only answer based on the provided PDF context.
• Secure Secrets: The .dockerignore and .gitignore files ensure your .env file is never leaked or baked into images.
• Performance Optimization: Streamlit's @st.cache_resource is used to prevent redundant loading of heavy ML models and database connections.

---

📄 License

Distributed under the MIT License.