HospitalShield

Inspiration

Hospitals depend on interconnected resources, including staffing, essential supplies, transportation, and backup power. A shortage in even one area can create operational challenges, especially during emergencies. I wanted to explore how AI could help identify potential vulnerabilities earlier and support better-informed contingency planning.

This idea inspired me to build HospitalShield, an AI-assisted hospital resilience prototype that connects resource monitoring, risk assessment, research, and contingency planning in one application.

What it does

HospitalShield uses simulated hospital data to identify resource constraints and generate contingency recommendations. Users can test scenarios such as driver shortages, nurse shortages, emergency kit shortages, and low backup generator fuel.

The application also retrieves external disruption reports as supplementary context and generates an AI-assisted decision report summarizing the simulated hospital's status, identified risks, proposed responses, and considerations for human review.

How we built it

I built HospitalShield using Python and Streamlit. I separated the application into modules for hospital simulation, resource assessment, contingency planning, external research, and AI-generated explanations.

The system uses predefined resource thresholds to identify simulated risks and create contingency plans. I integrated external research using Tavily and AI-generated reporting through Groq. The application presents proposed actions alongside validation requirements and human-review safeguards.

Challenges we ran into and what we learned

One of my biggest challenges was handling irrelevant external search results. A recent article is not necessarily relevant to a particular disruption, and an external report does not establish that a simulated hospital is affected. I worked on filtering research results and clarifying the distinction between external context and verified operational information.

I also learned the importance of modular software design, testing different scenarios, consolidating overlapping research reports, and separating deterministic risk checks from AI-generated explanations. This project showed me that building a useful AI application involves more than connecting an API and that it requires careful validation, transparent limitations, and safeguards against unsupported conclusions.

What's next for HospitalShield

Future improvements could include stronger source relevance checks, more comprehensive corroboration of external reports, expanded simulations, and clearer explanations of uncertainty. With appropriate validation and authorized integrations, a future version could explore how resilience planning might support real hospital operations.

HospitalShield is currently a prototype using simulated hospital data. It does not connect to real hospital systems, and its recommendations are proposals that require human review and approval.