Simple Retrieval-Augmented Generation (RAG) Agent

LIVE-DEMO:

https://simple-rag-agent.onrender.com

A simple Retrieval-Augmented Generation (RAG) agent that:

• Ingests PDF documents into a Chroma vector store
• Embeds text via Hugging Face Inference API (all-MiniLM-L6-v2)
• Retrieves relevant excerpts on user queries
• Generates contextualized answers using SeaLion LLM (e.g. Gemma-SEA-LION-v3-9B-IT)
• Exports responses as downloadable PDF files

Built with Python, Gradio, LangChain, ChromaDB, PyPDF2 & FPDF

---

Table of Contents

1. Features
2. Prerequisites
3. Installation
4. Configuration
5. Usage
6. Project Structure
7. API Reference
8. Contributing
9. License

---

Features

• PDF Ingestion: Splits each page into 1,000-character chunks (200-character overlap) for efficient indexing.
• Embeddings: Uses the Hugging Face Inference API endpoint for sentence-transformers/all-MiniLM-L6-v2 to generate vector embeddings.
• Similarity Search: Retrieves top-k most similar document chunks via ChromaDB.
• Contextual Q&A: Builds a prompt including retrieved context and queries SeaLion’s hosted LLM API for answers.
• Downloadable Reports: Packages answers (with listed source titles/pages) into a PDF for offline reference.

---

Prerequisites

• Python 3.8+
• Docker (optional, for running a local ChromaDB instance)
• A SeaLion API key (SEA_LION_API_KEY)
• A Hugging Face Inference API token (HUGGINGFACE_API_TOKEN)

---

Installation

1. Clone the repository

   git clone https://github.com/AlvesMH/RAG_agent.git
   cd simple-rag-agent

2. Create & activate a virtual environment

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

3. Install dependencies

   pip install -r requirements.txt

4. (Optional) Run ChromaDB in Docker

   docker run -d --name chroma -p 8000:8000 -v $(pwd)/chroma_db:/chroma chromadb/chromadb:latest

---

Configuration

Copy the example environment file and fill in your API keys:

cp .env.example .env
# Then open .env and add:
# SEA_LION_API_KEY=your_sealion_key
# HUGGINGFACE_API_TOKEN=your_hf_token

• SEA_LION_API_KEY: Used to authenticate calls to SeaLion’s chat/completions endpoint.
• HUGGINGFACE_API_TOKEN: Grants access to the Hugging Face Inference API for embeddings.

---

Usage

1. Launch the Gradio app

   python Simple_RAG_Agent.py

   This will open a local web interface (usually at http://localhost:7860).

2. Ingest PDFs

   • Click Upload PDF Documents → select one or more .pdf files.
   • Click Add to VectorDB → see ingestion status.
   • Check existing docs via List Documents.

3. Retrieve Excerpts

   • Enter a question in the Enter your question box.
   • Click Retrieve Excerpts to view top-k matching chunks.

4. Generate Full Response

   • With your query entered, click Generate Response.
   • The agent will call SeaLion, stitch context into an answer, display it, and save a PDF you can download.

---

Project Structure

├── .env.example            # Example env file
├── requirements.txt        # Python dependencies
├── Simple_RAG_Agent.py     # Main application script
├── chroma_db/              # Vectorstore
├── README.md               # This file
               

---

API Reference

SeaLion Chat Completion

• Endpoint: POST https://api.sea-lion.ai/v1/chat/completions
• Payload: { model, messages, temperature }
• Response: { choices: [ { message: { content } } ] }

Hugging Face Feature Extraction

• Endpoint: POST https://api-inference.huggingface.co/pipeline/feature-extraction/{model}
• Payload: { inputs: [...texts] }
• Response: [[float...], ...] embeddings

---

License

This project is distributed under the MIT License. See LICENSE for more details | Built with ❤️ in Singapore