A fictional insurance company processes incoming claims.
Today, every claim is manually reviewed.
This creates unnecessary manual work because many claims are low-risk and could move through the process automatically.
The goal is to build an executable workflow that:
- receives a claim;
- retrieves policy information from a legacy system;
- evaluates the claim risk;
- automatically routes low-risk claims;
- sends higher-risk or uncertain claims to human review;
- completes the process;
- records enough information to compare the baseline process with the improved process.
Every claim requires human review.
Claim Received
↓
Human Review
↓
Claim Completed
Claim Received
↓
Legacy Policy Lookup
↓
Risk Triage
↓
Decision
/ \
Low Risk Review Required
↓ ↓
Automatic Human Review
\ /
\ /
↓
Claim Completed
The workflow must call a simulated legacy policy system.
The legacy service will be implemented as a local FastAPI application.
At least one synthetic claim must deliberately trigger a legacy API failure.
The project must demonstrate:
failure
→ diagnosis
→ correction
→ retry
→ successful recovery
Claims that meet either of these conditions should require human review:
- elevated risk;
- low-confidence or uncertain automated assessment.
Compare:
All claims require human review.
Only claims requiring human judgment should be routed to human review.
The experiment should report:
- total claims;
- baseline manual reviews;
- improved manual reviews;
- reduction in manual review;
- failed integration attempts;
- recovered process instances.
- synthetic data only;
- local development environment;
- no production credentials;
- no production insurance data;
- workflow must remain understandable and reproducible;
- AI should assist the workflow, not obscure the business logic.
GSB-001 is successful if:
- the BPMN workflow deploys to Camunda;
- synthetic claims can be started as process instances;
- Python workers complete automated tasks;
- the legacy API integration works;
- a deliberate integration failure can be observed and recovered;
- human review is visible in the workflow;
- baseline and improved metrics can be compared;
- another developer can reproduce the build from GitHub.
This Signature Build follows the progression:
business process
↓
technical model
↓
working system
↓
observable failure
↓
measurable improvement
↓
customer handoff
The project should begin from the business requirement rather than from implementation code.
That means separating:
WHAT the business needs
from:
HOW we implement it
The purpose of GSB-001 is to demonstrate the ability to translate an ambiguous business process into an executable, observable, measurable, and reproducible technical system.