Project Story: InsuMAS

Inspiration

Insurance is often overwhelming, with complex policies, confusing claims, and slow support systems. We wanted to create an AI-powered solution that makes insurance simple, fast, and accessible for everyone. Inspired by the idea of teamwork, we designed a multi-agent system where each AI agent specialises in one part of the insurance journey — from providing information, to assisting with health problems, emotions, and insurance finance. The system also provides recommendations for insurance and doctor.

What We Learned

• How to build a multi-agent system which can divide complex problems into smaller and specialized tasks.
  
• The importance of user experience in making technical solutions approachable.
  
• How to integrate different AI models into one cohesive platform.
  
• Collaboration skills, since building with multiple agents required us to think like a team of agents ourselves.
  

How We Built It

1. Architecture: We divided the system into sub-models (agents). For example:

   • An Information Agent for answering questions related to insurance plans.
     
   • A Optimism Agent for supporting emotional-related issues.
     
   • A Insurance Recommend Agent for recommending appropriate insurance plans.
   • A Doctor Recommend Agent for recommending talented doctors
   • A FAQ Agent for answering general insurance-related questions
   • A Calculation Agent for estimating the amount remaining after insurance coverage

2. Tech Stack:

   • Backend: [Python, Numpy, Pandas, RAG].
     
   • AI: Multi-agent orchestration with [Langchain, Langgraph, Groq-API, Gemini-API].
     
   • Frontend: [Gradio] for a simple and user-friendly interface.
     

3. Integration: Each agent communicates with a central supervisor, making it feel like one smart assistant even though multiple models are working behind the scenes.

Challenges

• Orchestration: Getting multiple agents to work together smoothly was harder than expected.
  
• Context sharing: Passing the right information between agents without confusing the user.
  
• Time constraints: Balancing ambitious ideas with what could realistically be built during the hackathon.
  
• Insurance complexity: Simplifying policies into plain language without losing accuracy.
  

A Little Math

We even thought of our project like a system of functions:

[ \text{InsuranceHelp}(x) = f_{\text{info}}(x) + f_{\text{claims}}(x) + f_{\text{policy}}(x) ]

Where each ( f ) is a specialized agent, and together they form one complete solution.

Tagline: One assistant, many experts — smarter insurance with AI.

Built With

• communication
• data
• gemini-api
• gradio
• groq
• langchain
• langgraph
• multi-agent
• numpy
• pandas
• python
• rag