Medical Chat Assistant Backend

This backend service powers a multi-agent system designed to assist with medical queries by leveraging various research tools and language models. It consists of a Researcher Agent and a Research Manager Agent that work together to gather, compile, and respond to user queries with accurate and relevant medical information.

Overview

Multi-Agentic System

• Researcher Agent:

  • Receives the user's query and searches for relevant information across various sources, including PubMed, Arxiv, Tavily, and Google Scholar.

• Research Manager Agent:

  • Compiles the information retrieved by the Researcher Agent, draws conclusions, and frames the response in the language of a medical professional.

Tech Stack

• LangChain: A framework for integrating large language models (LLMs) with web scraping tools.
• CrewAI: A library that facilitates the use of multi-agent Retrieval-Augmented Generation (RAG) systems.
• Flask: A Python-based web framework used to host the chatbot backend.
• Gemini-1.5-Flash: The LLM used for generating responses within the chatbot.
• LangChain Tools:
  • DuckDuckGo Search
  • Tavily Search
  • Semantic Scholar
  • Google Scholar
  • Arxiv
  • Serper Dev Tool

API Endpoints

1. Initiate a Research Query

• Endpoint: /api/crew
• Method: POST
• Request Body:

  {
    "question": "user query related to his medical condition..."
  }

  This endpoint starts the multi-agent system with the given query. The process will run asynchronously, and a jobid will be returned to track the progress.
• Response:

  {
    "jobid": <The ID of the subprocess handling the query.>
  }

2. Check the Status of a Research Query

• Endpoint: /api/crew/<job_id>
• Method: GET
• Request Body:

  {
    "job_id": <The ID of the subprocess you want to check.>
  }

This endpoint is used to poll the status of the initiated query. If the process is completed, the response from the AI system will be returned.

• Response:
  • The status of the query (e.g., processing, completed).
  • If completed, the final response generated by the multi-agent system.