🤖 Modular RAG Chatbot

An production-grade, stateful RAG (Retrieval-Augmented Generation) chatbot built with LangChain, Streamlit, and ChromaDB. Designed for high performance, local privacy, and $0 cost.

🌟 Key Features

• Conversational Memory: Understands flow and follow-up questions using history-aware retrieval.
• Source Attribution: Cites specific PDF page numbers in answers.
• Local Vectors: Embeddings (HuggingFace) and database (ChromaDB) run 100% locally.
• Production Standards: Modular architecture, Google-style docstrings, and robust error handling.
• Inspector Tool: Built-in tab to visualize internal vector chunks and metadata.

🏗️ Architecture

graph TD
    A[PDF Upload] --> B[Document Processor]
    B --> C[Vector Store Manager]
    C --> D[(Local ChromaDB)]
    E[User Query] --> F[RAG Engine]
    D --> F
    F --> G[Gemini 2.0 LLM]
    G --> H[Response with Sources]

Loading

🚀 Quick Start

1. Prerequisites

• Python
• A Google Gemini API Key (Get it here)

2. Setup

# Clone the repository
git clone <your-repo-link>
cd <your-directory>

# Create and activate virtual environment
python -m venv venv
.\venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

3. Environment Variables

Create a .env file in the root directory:

GOOGLE_API_KEY=your_key_here

4. Run the App

streamlit run app.py

📂 Project Structure

• app.py: Main entry point.
• ui_components.py: Modular Streamlit UI logic.
• document_processor.py: PDF loading and chunking.
• vector_store_manager.py: ChromaDB lifecycle management.
• rag_engine.py: LangChain pipeline and memory logic.

---

Created for Workshop Demonstrations - Distributed under the MIT License.