Cross-Sight

• 

  Simulated pollution plume and affected hospitals.

Inspiration London’s Great Smog of 1952 killed thousands because hospitals had no warning. That pattern still happens today: pollution and heat drive respiratory surges days later, but hospitals only react once beds fill. We built Crosssight to close that gap.

What it does Crosssight is a live respiratory early-warning console for London hospitals. It pulls live air quality and weather, scores each catchment’s respiratory pressure, shows it on a 3D map, and uses an AI agent to explain the risk, draft a readiness alert, and on approval email the supervisor and reach out to high-risk patients in the catchment.

How we built it We built a FastAPI backend with live LAQN and Met Office feeds, a deterministic risk engine grounded in the hackathon’s System climate research graph, and Claude agents for reasoning and alerts. The React/Cesium frontend consumes that backend end to end. Patient severity comes from Apollo, vulnerability from Milliman SVI-style indices mapped to each catchment.

Challenges we ran into The clinical datasets couldn’t support a pollution-to-admissions ML model, so we had to be honest and build on published evidence instead. Wiring the frontend to the backend, handling live API failures gracefully, and getting real email dispatch working took more iteration than we expected.

Accomplishments that we're proud of We shipped a full pipeline from live environmental data to real emails in the inbox. Every risk score is traceable to cited studies. The demo climax works: simulate an episode, watch the agent reason, click Agent ACT, and two emails land. We refused to fake a predictor we couldn’t defend.

What we learned Transparent arithmetic plus agentic reasoning beats a black-box model when your data can’t backtest. Judges and clinicians care more about honesty and lead time than a flashy accuracy number. Prevention needs operational tooling, not just another dashboard.

What's next for Cross-Sight Swap Milliman California vulnerability weights for GLA borough data, plug in live UKHSA alerts, and pilot with an NHS trust using lead time and admissions as the primary KPI. Longer term: NHS Spine integration for real patient outreach and a stepped-wedge evaluation across hospital catchments.

Built With

• cesium
• cesium-ion
• data
• google-photorealistic-3d-tiles-other:-gmail-smtp-(email-dispatch)
• json/file-based
• languages:-python
• met-office-weather-api
• pydantic
• server-sent-events-(live-agent-stream)
• typescript-frontend:-react
• uvicorn-ai:-anthropic-claude-(sonnet-4.6)-apis-&-data:-laqn-london-air
• vite
• zustand-backend:-fastapi