Inspiration

The insurance industry faces complex operational challenges, including huge volume of document ingestion from various sources, manual document classification, fragmented and manual claim verification, and lack of centralized knowledge management. Insurers drown in paperwork, with time-consuming and error-prone scanning, uploading, and categorization of claim dockets. Claim Adjuster navigate through disparate systems (Guidewire, Document Management System, etc..) and handwritten forms, hindering access to right information for timely and accurate claim verifications

What it does

Our solution offers a powerful answer to the operational challenges faced by the insurers, streamlining the ingestion from various sources, enhancing the arduous process of document classification, indexing and verification with exceptional precision and speed. It can also seamlessly integrate with insurers systems to improve compliance checks and adherence to regulations.

How we built it

Discussed with Domain SMEs to put together Industry challenges Use Cases, prepared the Process Flow and Technical Architecture. Initially the thought process was to have an Orchestrator Agent which will decide which sub-agent to call based on the outcome of each agent. Since we were facing issue in extraction Agent (due to huge document size), we changed it into a Container Image Lambda and followed the architecture that is attached to this solution.

Challenges we ran into

Extraction of Huge Documents (Structures, Unstructured) was taking time to extract resulting in Agent was not responding.

Accomplishments that we're proud of

Asynchronously documents are getting ingested from various sources (email, scan and upload, etc..) Automatic Classification, Translation, Entity Extraction, then based on confidence score threshold automatically updating the entities into Guidewire and DMS system or sending to Human In The Loop (HITL). Legal Scenario is the most happening Use Case in the industry as it automatically identifies the legal case documents are sensitive or non-sensitive and does a handshake with Guidewire & DMS system by updating legal entities.

What's next for HCLTech Claim Assist

Next step is to use make individual lambda features as MCP Tools and using AgentCore Gateway those tools shall be used. Also using AgentCore Runtime, deploy the Agents (Strand Agents).

Note: The demo videos are placed in github, they are not uploaded on youtube due to restriction.

Built With

  • apigateway
  • bedrock
  • bedrockguardrails
  • cloudtrail
  • cloudwatch
  • cognito
  • dynamodb
  • ecr
  • ecs
  • iam
  • lambda
  • llm
  • react
  • s3
  • secretmanager
  • sns
  • sqs
  • textract
Share this project:

Updates