Weekend Creative Agent Challenge

Know which door to knock on first.

LeadSentry turns an address into an explainable childhood lead-exposure risk assessment, helping public-health teams direct testing and outreach where limited resources can matter most.

The mission

Prevention is an allocation problem.

Health departments can test for lead exposure, distribute filters, and conduct outreach. The hard part is deciding which addresses to prioritize before harm is detected. LeadSentry combines environmental evidence, housing-age context, and explicit confidence rules into one traceable decision.

Not a diagnosis. LeadSentry is a transparent triage assistant. It ranks where preventive attention may be most valuable while preserving the sources, missing inputs, and location-quality caveats behind every result.

0–100Deterministic risk score
26Erie County ZIPs validated
0.55Land-weighted mean Spearman correlation

AWS architecture

One request. Five accountable steps.

The model does not invent the score. Deterministic tools own calculation; the agent owns interpretation. AWS supplies a repeatable, serverless path from request to persisted assessment.

01 / REQUEST Client

Sends an illustrative address as JSON over HTTPS.

02 / ROUTE Function URL / API Gateway

Expose public HTTPS entry points backed by the same Lambda.

03 / EXECUTE AWS Lambda

Validates input and runs the existing LeadSentry workflow.

04 / SCORE Risk engine

Combines housing age, contamination, and water-service evidence.

05 / RECORD DynamoDB

Stores the full assessment, evidence, action, and timestamp.

Pre-1980 housing 50

US Census ACS B25034 joined through the Mireye tract GEOID.

Legacy contamination 30

EPA Superfund, brownfield, RCRA, and leaking-tank proximity signals.

Water service gap 20

Community water coverage and domestic-well density evidence.

Live on AWS

Exercise the deployed scoring workflow.

Choose a documented fixture and invoke the public Lambda Function URL. The response is calculated live and persisted to DynamoDB. All fixture values are clearly labeled as illustrative.

Request

Select a demo address

Each option exercises a different risk profile in the illustrative dataset.

Public demo only. These environmental and housing values are invented fixtures shaped like the production data contract.

Assessment

Ready for a live request

Select a fixture and run the workflow to see its score, confidence, action, and cited components.

Ground-truth check

A ranking should meet reality.

Across 26 Erie County ZIP codes, LeadSentry's land-weighted mean score was compared with public childhood blood-lead outcomes from the New York State Department of Health.

Correlation does not turn a triage score into a clinical diagnosis. It does provide encouraging evidence that the ranking captures a useful signal for deciding where deeper investigation may be warranted.

ρ 0.55Spearman rank correlation
r 0.44Pearson linear correlation
Scatter plot comparing LeadSentry ZIP risk scores with observed elevated blood-lead rates in Erie County
Higher LeadSentry risk scores generally align with higher observed elevated-BLL rates per 1,000 children tested.

Built for accountable action

An agent should show its work.

LeadSentry preserves source provenance, exposes missing inputs, compares agent decisions with a rule baseline, and keeps its deterministic scorer testable. The result is not a black-box prediction; it is an auditable recommendation for where preventive attention may matter.