Multi-Agent-System-for-Healthcare-Appointment-Booking

**Multi-Agent System for Healthcare Appointment Booking**
---

A modular, extensible framework for managing healthcare appointment workflows using multiple AI agents, including:

- **Router Agent**: Directs user requests to specialized services (new bookings, cancellations, general inquiries).
- **RAG Agent**: Provides question-answering capabilities by retrieving context from PDF documents.
- **Tool Nodes**: Interfaces for booking, cancellation, doctor search, and availability checks.
- **Audio Output**: Converts AI responses to speech via ElevenLabs.

---

## State Transitions

This GIF illustrates how the router inspects message history and transitions between nodes:

![State Transitions](https://github.com/JHansiduYapa/Multi-Agent-System-for-Healthcare-Appointment-Booking/blob/main/src/short_video.gif)

---

## Features

1. **Custom Router**
   - Implements routing logic from scratch.
   - Routes user queries to appropriate agent nodes based on intent.
   - **Router Implementation**: Built with LangGraph’s command functions, it inspects the full message history at each turn to decide the next node (e.g., booking, cancellation, RAG or audio) and dynamically dispatches the conversation to the correct assistant.
   - **Router Code**:

     ```python
     def router_model(
         state: State
     ) -> Command[Literal[
         "cancel_booking_assistant",
         "new_booking_assistant",
         "general_hospital_assistant",
         "audio_output",
         END
     ]]:
         """Route to correct worker agent. You only route the conversation."""
         llm_router = _get_llm().bind_tools([])
         llm_router_with_tools = llm_router.bind_tools([
             cancel_booking_assistant,
             new_booking_assistant,
             general_hospital_assistant
         ])

         custom_prompt = {
             "role": "system",
             "content": (
                 "You are a routing assistant. Based on the message history, "
                 "decide