Document Router

A generic document-based AI router that passes entire documents to specialized agents without RAG chunking.

Key Concept

• Full Document Context: Load complete PDF/document content into agent prompts (no vector databases or chunking)
• Tool-Based Routing: Agents use function calls to switch between each other based on question type
• Specialized Expertise: Each agent has access to specific documents for deep technical understanding
• Structured Outputs: Pydantic validation for clean agent switching and tool calls

Architecture

• DocumentRouter: Core routing system that manages agents and documents
• GeneralAgent: Routes general questions to appropriate document experts
• DocumentExpertAgent: Specialized agents with full document context
• Tool-based switching between agents using OpenAI function calling

Usage

# Basic setup
client = AzureOpenAI(endpoint="...", api_key="...", api_version="...")
router = create_document_chatbot(client, config)

# Interactive chat
router.chat("How does the authentication work?")  # Routes to appropriate expert

Installation

uv sync
uv run python main.py

Files

• main.py: Original implementation (now generic)
• generic_document_router.py: Reusable router framework
• example_config.py: Configuration examples for different use cases