Pediatric Fever Chatbot API

Welcome to the documentation for the Fever Model of Docokids. This system uses a state-of-the-art LangGraph agent to perform pediatric fever triage based on international clinical guidelines.

Key Features

  • Clinical Triage Graph: A structured state machine (LangGraph) that guides the conversation from data collection to recommendation.
  • Urgency Detection: Real-time analysis of red flags (e.g., convulsions, breathing difficulty) and risk factors (age < 3 months).
  • Checklist-Based: Ensures all critical information (age, temperature, duration, hydration) is collected before making a recommendation.
  • FastAPI Integration: Secure, high-performance API with streaming support.
  • PostgreSQL Persistence: Stores conversation state and user feedback.

Architecture

The system is built on LangGraph, where the conversation flow is modeled as a graph of nodes:

  1. Receptor: Extracts structured information (e.g., “39°C”, “2 years”) from user messages.
  2. Triage Route: Analyzes the state to decide the next step:
    • Urgency Recommendation: Immediate advice if red flags are detected.
    • Recommendation: Final advice if enough information is gathered.
    • Inquiry: Ask follow-up questions if data is missing.
  3. Nodes: specialized Python functions that handle each step.

Tech Stack

  • Framework: FastAPI
  • Agent Orchestration: LangGraph / LangChain
  • Database: PostgreSQL (psycopg3)
  • Runtime: Python 3.12+
  • Testing: Pytest

Getting Started

Check out the Getting Started guide to set up the project locally.

API Reference

Explore the endpoints in the API Reference.