TwinHospital

Inspiration

• The idea of a "digital twin" for hospitals — a live, data-driven simulation that mirrors reality and lets teams test changes safely — felt like a natural way to help clinicians, administrators, and engineers make better decisions faster.
• We wanted a tool that lets teams explore "what if" scenarios without disrupting care, that surfaces bottlenecks, and that can be extended to support planning, training, and research.

What it does

TwinHospital models key hospital systems (triage, wards, ED, imaging, and operating theaters) and simulates patient flows, resource usage, and staff scheduling. It provides:

• A simulation engine that runs scenarios (normal day, surge, staffing shortage) and outputs measurable KPIs (throughput, wait time, bed occupancy).
• An interactive visualization layer to explore floor plans, queues, and timelines in near real-time.
• Scenario configuration and comparison so users can test interventions (add a nurse, re-route patients, open a surge ward) and compare outcomes side-by-side.
• Exportable reports and telemetry for post-scenario analysis and auditing.

Challenges we ran into

• Modeling fidelity vs. simplicity: Hospital systems are full of edge cases. Early models were too simple to reflect important bottlenecks; increasing fidelity risked overfitting to one hospital’s processes. We struck a balance by making core behaviors parameter-driven so users can tune complexity.
• Usability across roles: Clinicians, administrators, and engineers have different mental models. Designing a UI and scenario workflow that serves all three required repeated user testing and simplifying the language and controls.

Accomplishments that we're proud of

• A reusable simulation core that can model multiple hospital areas and supports scenario comparison out of the box.
• Intuitive visualizations that helped non-technical stakeholders identify a simple staffing change that reduced average ED wait time in early user tests.

What we learned

• Domain knowledge is essential: Without clinician and operations input, simulations can produce plausible but meaningless results. Close collaboration with end users dramatically improves model relevance.
• Start simple and make complexity configurable: Provide sensible defaults and expose knobs for advanced users rather than forcing complexity on everyone.

What's next for TwinHospital

• ML-driven predictions: Integrate short-term forecasting models (admissions, length-of-stay) to make scenarios more predictive.
• Real-time ingestion: Add secure connectors for streaming telemetry (EHR, patient flow trackers) to support near-real-time digital twin operation.

Built With

• ai-agent
• python