FairQuote

Know your price and the reason behind it. Insurance made transparent

Comment

FairQuote: Intelligent Underwriting for Car Insurance

Inspired by: McKinsey – The Future of AI in Insurance

Goal: Build a functional, agentic AI system for insurance underwriting that is demo-ready.

Inspiration

Traditional insurance underwriting is slow, opaque, and manual. Users fill forms repeatedly, document validation takes days, and premium calculations are not transparent.

We wanted to demonstrate how AI can accelerate underwriting and improve user experience, while keeping it realistic and functional.

What it Does

FairQuote provides an AI-driven conversational experience for insurance underwriting:

  • Conversational data collection with intelligent intake
  • Risk scoring and transparent premium calculation
  • Policy document generation with plain-English explanations
  • Real-time vehicle validation via a mock DVLA API

All of this runs in a single conversation with three streamlined agents orchestrated by an AI-driven Orchestrator.

How We Built It

Hackathon-Friendly Architecture

Simplified 3-Agent System + Orchestrator

ORCHESTRATOR AGENT
├── Routes requests to specialized agents
├── Maintains session state

    ├── INTELLIGENT INTAKE AGENT 💬
    ├── RISK & PRICING ENGINE 🧮
    └── POLICY DOCUMENT GENERATOR 📄

Agent 1: Intelligent Intake Agent 💬

"Conversational Front Door"

What it Does:

  • Collects user and vehicle info naturally
  • Handles document uploads (driving license OCR)
  • Validates vehicle via mock DVLA JSON
  • Detects inconsistencies in user input

Tools Used:

  • Claude API (Sonnet 4.5) for conversation & function calling
  • Python functions for validation
  • Mock DVLA data for demo purposes
  • Optional OCR via Textract (or placeholder for demo)

Agent 2: Risk & Pricing Engine 🧮

"The Decision Brain"

What it Does:

  • Computes risk score using simple formulas
  • Calculates premium with transparent breakdown
  • Generates explanations and recommendations

Tools Used:

  • Python formulas for risk/premium
  • Claude API (Sonnet 4.5) for natural language explanations
  • Simple RAG system with Chroma (50 risk explanation snippets)

Agent 3: Policy Document Generator 📄

"Legal Assembler"

What it Does:

  • RAG-based retrieval of policy clauses
  • Assembles full policy document
  • Generates plain-English summaries

Tools Used:

  • Claude API (Sonnet 4.5) for document assembly
  • Chroma vector DB with 50–100 policy clause snippets

Orchestrator Agent 🎯

  • Routes conversation across the three agents
  • Maintains session memory
  • Ensures workflow: Intake → Risk & Pricing → Policy

Challenges We Ran Into

  • Limited demo datasets for vehicle validation and OCR
  • Small RAG datasets needed for hackathon-friendly explanations
  • Balancing realism vs. demo-readiness

Accomplishments We’re Proud Of

  • Fully functional end-to-end AI underwriting demo
  • Agents are modular and easily extendable to real services
  • Transparent risk scoring and premium calculation workflow

What We Learned

  • Multi-agent orchestration simplifies complex workflows
  • Claude (Sonnet 4.5) is powerful for both conversation and document assembly
  • Hackathon constraints force creative use of mock data for realistic demos

What’s Next for FairQuote

  • Integrate real DVLA and OCR services
  • Expand RAG datasets for richer explanations
  • Add multi-language and voice support
  • Enhance policy document generation with dynamic templates

Built With

  • aws-agentcore-memory
  • aws-bedrock
  • aws-bedrock-knowledge-base
  • aws-cloudwatch
  • aws-iam
  • aws-lambda
  • aws-strands-sdk
  • landingai-agentic-document-extraction
  • rag
Share this project:

Updates