Case study 03
Agile Underwriting / AI Automation
From unstructured information to intelligent workflows.
- Client
- Agile Underwriting
- Engagement
- Delivered through Bilue
- Discipline
- Applied AI, AWS, Intelligent Workflows
- Stack
- AWS serverless, Amazon Bedrock, Claude, AWS CDK
01
Overview
I led the design and delivery of an AI platform that automates cyber insurance quote processing and broker email triage. It is built serverless on AWS, with Amazon Bedrock and Claude doing the reading and deciding.
02
Challenge
Quote requests arrive as broker emails: unstructured, inconsistent and full of attachments. Getting from an email to a quote means finding the right facts, checking them, and knowing when a person needs to look.
03
Responsibilities
- Led the design and delivery of the platform.
- Implemented the agentic email-processing loop on Amazon Bedrock, where Claude either submits a structured quote or flags the email for human review.
- Defined the infrastructure as code with AWS CDK, with typed configuration per environment.
04
Architecture
Each email enters an agent loop on Amazon Bedrock. Claude works through the email using tools and ends in one of two ways: a structured quote is submitted, or the email is flagged for a person to review.
05
Decisions
Two exits, both explicit
The agent has exactly two ways to finish: submit a structured quote, or flag for review.
Structured output, not prose
A quote leaves the loop as structured data that downstream systems can check.
Typed configuration per environment
The CDK infrastructure shares typed configuration across environments.
06
Trade-offs
07
Results
08