Operators in 911 call centers must rapidly understand a caller’s situation, assess urgency, and route the patient to the right hospital, often with incomplete information and under intense time pressure.
The scope of a rapid design sprint allowed us to explore different ways to help emergency operators make faster, more accurate, and trustworthy decisions with AI, while preserving human control and accountability.
Staffing shortages: Public Safety Answering Points (PSAPs) struggle with understaffing and high turnover due to constant multitasking and high-stress environments.
Legacy limitations: Legacy systems cannot handle modern data like video or IoT feeds, forcing dispatchers to rely on verbal descriptions that are difficult to manage in high-stress or multilingual situations.
Detection sensitivity assistance: Machine learning models listening to live calls have a significantly higher sensitivity for recognizing cardiac arrest (85.0%) compared to dispatchers alone (77.5%). Models can also assist with transcription and translation across language barriers in diverse communitites.
Reduce ER / transport-to-treatment time while maintaining accuracy.
Increase routing accuracy under operational constraints.
Increase ticket efficiency by reducing cognitive load.
All-in-one emergency response hub
AI assisted ambulance routing
Automated patient document summary
Human validation
While not a complete implementation, these screens illustrate critical interaction patterns:
Intake Dashboard: Structured capture of patient info and immediate risk indicators
Hospital Recommendation Panel: Suggested hospital, confidence score, and explanation
Human-in-the-Loop Controls: Edit case information, add missed routing opportunities
One-click Communication: Synchronized info sent to ambulance and hospital
The AI transcribes the emergency call live. The human operator can validate this information later.
The AI processes the transcript and extracts key entities from the live call transcription (symptoms, age, risk factors.) If critical information is missing, it notes a flag for the operator to send to the the EMT and hospital.
Outcome: A structured patient profile ready for assessment.
The system combines clinical urgency with real-time operational data: travel time to nearby hospitals, ER wait times / queue load, staff coverage, and bed availability. It then ranks hospitals based on best overall fit, not just proximity.
Outcome: A prioritized hospital recommendation list.
After the human operator validates or overrides the decision, the system logs and synthesizes patterns for input, AI recommendations, and operator decisions.
Outcome: Data that is used to audit decisions and improve future model performance.
While this project isn’t a finished product, it explores how UX design integrates with ethics, operations, and AI transparency in environments where every second counts. Effective AI decision support emerges when product clarity, data reliability, and cross‑functional collaboration work in sync.
Working with a multidisciplinary team required translating and reinforcing concepts across perspectives to ensure true alignment and shared understanding
Clear scoping and restraint is required for time-bound projects
High-stakes AI trust depends on reliable inputs and transparency
AI–human interaction patterns should focus on transparency and decision support while aligning with operational constraints
Integration with real health systems data
Expanded explainability and alternative paths exploration
Operator training simulations
Edge‑case error handling (low AI confidence or missing data)
More features: weekly intake summary, intake for videos & texts in addition to landline call
Pleihouse—Design system for a playful e‑commerce brand.
Lululemon Audit & Redesign—Improving discoverability and personalization for the Lululemon experience.