RiskWeave

Application form
Application form
Home Page
Underwriting decision and the stage at which it got rejected

Inspiration

Underwriting across domains—finance, medical, behavioral—is often siloed, slow, and lacks intelligent risk assessment. We envisioned RiskWeave as a unified agentic system that could intelligently weave together diverse data sources to assess risk in real time, using LLMs and graph-based reasoning.

What It Does

RiskWeave is an AI-powered underwriting engine built on AWS that:

Uses agentic architecture with specialized agents (Risk, Audit, Logging, NLP).
Integrates banking, medical, and behavioral data from S3 folders.
Employs LLMs via Amazon Bedrock for document understanding and reasoning.
Leverages a graph database to connect entities and infer risk patterns.
Provides real-time, explainable risk scores and audit trails.

How We Built It

Agents were built using Bedrock and Agent Core
Data ingestion is sent S3 was orchestrated using FastAPI and Boto3 sdk
LLMs (Claude) were accessed via Amazon Bedrock for NLP tasks.
Graph reasoning was implemented using neo4j and AWS Bedrock KB
Audit and logging agents used CloudWatch and OpenSearch for traceability.
The system was containerized using Docker.

Challenges We Ran Into

Harmonizing data formats across domains (medical PDFs vs banking CSVs).
Ensuring LLMs could extract numerical figures reliably from unstructured text.
Designing agents that could collaborate without becoming tightly coupled.
Managing costs and latency during LLM inference at scale.

Accomplishments That We're Proud Of

Built a fully functional agentic underwriting system in under two weeks.
Achieved cross-domain risk scoring with explainability.
Enabled real-time audit logging and traceable decision paths.
Created a modular architecture that can be extended to new domains easily.

What We Learned

Agentic systems require clear role definitions and robust communication protocols.
LLMs are powerful but need guardrails for reliability in financial contexts.
Graph databases are ideal for entity resolution and relationship mapping.
AWS services like Bedrock and Agent Core can be seamlessly integrated for GenAI use cases.