RAG Chatbot¶

Welcome to the RAG Chatbot documentation. This guide covers installation, configuration, and usage of the Retrieval-Augmented Generation (RAG) chatbot system.

Overview¶

RAG Chatbot is an intelligent document-based chat system that allows users to ask questions and receive accurate answers based on your uploaded content. It combines the power of large language models (LLMs) with a searchable knowledge base built from your documents.

Key Features¶

• Intelligent Q&A - Ask questions in natural language and get answers sourced from your documents
• Multiple LLM Providers - Choose between OpenAI (GPT-5.1) or Anthropic (Claude) for chat responses
• Document Support - Upload PDFs, Word documents, PowerPoints, spreadsheets, images, and audio files
• Media Import - Import transcripts from YouTube, Vimeo, SoundCloud, and 1000+ other platforms
• Hybrid Search - Combines semantic (AI) search with keyword matching for accurate results
• Conversation Memory - Maintains context across long conversations with automatic summarization
• Embeddable Widget - Drop-in chat widget for any website

How It Works¶

┌─────────────┐     ┌──────────────┐     ┌─────────────┐
│   Upload    │────▶│   Process    │────▶│   Store     │
│  Documents  │     │  & Chunk     │     │  Vectors    │
└─────────────┘     └──────────────┘     └─────────────┘
                                                │
                                                ▼
┌─────────────┐     ┌──────────────┐     ┌─────────────┐
│   Answer    │◀────│   Retrieve   │◀────│   Query     │
│   User      │     │   Context    │     │   Database  │
└─────────────┘     └──────────────┘     └─────────────┘

1. Upload - Documents are uploaded through the web interface or API
2. Process - The Docling service parses documents and extracts text
3. Store - Text is split into chunks and stored with vector embeddings
4. Query - User questions are converted to vectors and matched against stored content
5. Retrieve - The most relevant chunks are retrieved using hybrid search
6. Answer - An LLM generates a response based on the retrieved context

System Requirements¶

Supported Platforms¶

• Operating System: AlmaLinux 9, RHEL 9, Rocky Linux 9, Ubuntu 22.04+
• Cloud: AWS EC2, Google Cloud Compute Engine, Azure VMs
• Container: Docker, Podman

Quick Start¶

For a quick installation, follow these steps:

1. Install the system requirements on AlmaLinux 9
2. Set up PostgreSQL 16 with pgvector
3. Configure the application
4. Deploy the Docling service
5. Start using the chat interface

Architecture¶

The system consists of three main components:

PHP Application¶

The main web application handles:

• Chat API endpoints
• Document upload and management
• User session management
• Communication with LLM providers

PostgreSQL Database¶

Stores all persistent data:

• Document metadata and content chunks
• Vector embeddings for semantic search
• Chat session history
• Full-text search indexes

Docling Service¶

A Python microservice that handles:

• Document parsing (PDF, DOCX, images, etc.)
• OCR for scanned documents
• Table and layout extraction
• Media downloading and transcription

Support¶

If you encounter issues:

1. Check the Troubleshooting guide
2. Review application logs in /var/www/chatbot/logs/
3. Use the Debug RAG endpoint to diagnose search issues

Next Steps¶

Ready to get started? Head to the Installation Guide to set up the system on AlmaLinux 9.