🏆 Winning the largest AI Hackathon in the World
Dispatch was initially built out of a spontaneous trip to the Berkeley AI Hackathon , where my team won the Grand Prize with $50,000 in investment from Skydeck, and an additional $14,000 in credits from Intel and OpenAI. Post-Hackathon, we decided to take this project further and launched a venture with Skydeck, the #2 top ranked Startup Accelerator in the US. I worked as the solo designer.
The Problem
💬 “How might we empower dispatchers to efficiently manage non-emergency calls in order to increase confidence in addressing critical situations and safeguard the caller's well-being?”
🚨 911 Dispatchers are Overwhelmed
911 dispatchers are overwhelmed with non-emergency calls, which account for 80% of all calls . This leads to long wait times for emergency calls, and inefficient resource allocation . Dispatch aims to solve this problem by automating the process of handling non-emergency calls.
🚨 Heart Attack Victim waits 62 Seconds for his call.
The Oakland and LAPD Dispatch have wait times that fail to meet government safety standards almost every day of the year.
Secondary Research
A Life or Death Situation
Imagine a major earthquake striking a densely populated city. Within seconds, thousands of frantic 911 calls flood the emergency call center. Every line is busy, every second counts, but the overwhelmed operators can't answer every call. This isn't just a hypothetical scenario—it's a grim reality for many emergency services today.
Understanding the User
Visiting the LAPD: the Highest-Response Time Dispatch Center
Driving 60 miles to visit the LAPD
When we got invited to tour the LAPD for an onsite tour, I decided to drive 60 miles to Culver City to talk with users IRL.
Speaking with 10+ Dispatchers and Deputy Chiefs
We plugged into 911 calls to observe procedures firsthand and also spoke with a Staff Psychologist about common dispatcher stressors.
Documenting Findings
After the meeting, I transferred my notes to Figjam and identified key workflows and behavioral patterns that would inform the design solution.
“We get a lot of non-emergency calls that distract us from critical calls.”
— LAPD Deputy Chief
Translating findings from the LAPD
After the visit, I connected the dots across data points, stakeholders affected, and pain points to understand the root cause and affected parties at the LAPD Center.
🔍 Design Opportunity
The LAPD dispatch center is
overloaded with non-emergency calls that exacerbate
their staffing shortage (which is uncontrollable), reducing the
quality of emergency response.
Offloading non-critical calls to an AI system could
reduce cognitive load and improve response times with the
same number of dispatchers
Uncovering Patterns
Our conversations with dispatchers, LAPD staff, and previous owners helped us uncover critical problem spaces to tackle. With this, we aimed to identify intervention points where DispatchAI could make the most impact.
Research Consolidation
Defining the Core User
After consolidating research from expert interviews, field / observational research from the LAPD, and conversations with the staff psychologist, we defined the core frustrations, experience, and aspirations of our target user: John.
Operator Flow
I developed a user flow after our conversation with the LAPD to map out key decision points (currently 3 in total marked by diamonds). The Current operator flow is messy with multiple manual steps and decisions that increase cognitive load . Our goal is to shift this paradigm to give operators back their confidence.
User Journey Mapping
This journey map visualizes the steps taken by operators during a 911 call, from the moment a call is received to the dispatch of emergency services. It highlights the pain points and opportunities for improvement in the current system.
🔍 Design Opportunity
The research revealed that poorly-documented emergency calls caused by the unpredictable circumstances of dispatch is the root cause of distress. The opportunity, therefore, is offloading documentation to a more accurate and efficient agent
Primary Pain Points
Design Strategy
Investigating Need and Priorities
Based on our characterization of the target user, we conclude the following being crucial considerations in our design strategy:
- Maintaining familiar workflows and tools from legacy systems
- Identifying the key touch points that empower decision-making
Analyzing the Threat of Communication Breakdowns
Communications Failure is consistently identified as the primary cause for delayed first-responder response. This has caused unnecessary and tragic deaths that could have been avoidable with more effective communication systems
Addressing the Communications Failure
To address the communication failures that delay first-responder actions, I developed flow diagrams mapping the critical roles and decision points during a live emergency. By identifying key dependencies and breakdowns, I used this to inform the design of the handoff flow, ensuring vital information is funneled to all incident stakeholders.
Structure based on the Incident Command System under NIMS (National Incident Management System) for standardized emergency response.
🎯 The Primary Objective
The primary goal is to increase interoperability and efficient handoff of critical information while reducing the cognitive load of the dispatcher. This requires a careful balancing act between automation and human-in-the-loop decision-making. This will be addressed in future architecture designs.
Prioritizing Features
With key decision points, operator frustrations, and the user
journey defined, we prioritized features that
directly addressed the core problem of cognitive overload
from nonemergency calls.
Mapping Key Actions and Decision Points with Human-in-the-Loop
Based on the prioritized features, we worked to map each core action on a timeline. We emphasized areas to incorporate humans in the decision-making process as to maintain dispatcher’s autonomy and expertise in the end-to-end emergency cycle.
Generating Scenarios for Edge Cases
Scenario Generation highlighted different categories of emergencies and mitigation strategies to address them with the DispatchAI platform.
KEY INSIGHTS
⏰ Non-Emergency Responses causes critical delays
Research demonstrated that 9/10 calls are noncritical. This is not only exacerbated by the staffing shortage, but causes critical mental health problems for dispatchers. In large-scale crises, delays can result in preventable deaths.
📡 Breakdown of Communication Systems Exacerbate Emergencies
Most calls are built around radio frequency. Interference with these brittle mediums can cause the entire communication line to collapse. This is what happened during the 9/11 attacks when hundreds of firefighter lives were lost due to insufficient communications infrastructure.
🧠 Operator Cognitive Load caused Major Errors
Manual instruction scripts, muffled calls, and quasi geolocation services are daily experiences for dispatchers. This has led to misinformed decisions, inaccurate location identification, and extreme stress for dispatchers.
Design Iterations
Defining Functional Requirements
On top of Product Design, I also acted as our team's Product Manager to translate and define requirements based on user insight. Our product goal is to design intuitive and customizable modules to enable dispatchers to make quicker decisions with more confidence and less stress.
With my team, we defined several core functional requirements:
- Clear alert system for handoff between the AI and Human Operator
- Incident management panel with real-time action suggestions
- Modular and customizable panels for situational awareness
The Dispatch AI Operational Loop
Dispatch AI is designed around a continuous operational loop where data is collected, analyzed, and integrated to inform real-time decision making. Data is logged and used to inform future decisions.
Low Fidelity Mockups
Based on feedback from the LAPD, the lofi design prioritizes
modularity through panel-based widget design and
meaningful organization of information so operators
can quickly access information that drive action.
The following were developed at this stage:
- Modular panels
- Modals and toasts
- Improving and building on the Hackathon MVP
High Fidelity Mockups
After finalizing the overall information architecture, I developed a design system with colors and icons aligned with the NATO Symbology guidelines for military and government applications. The modules were further refined with additional detail to align with the multi-tasking workflow dispatchers are familiar with. Fully equipped screens were prioritized over limiting information.
Final Design
View emergencies in real-time
Translate languages, view transcript, and approve recommendations during a live call.
Reviewing and approving action recommendations
AI recommends actions based on historic and live data while the operator reviews and selects those to approve.
View previous calls to make decisions
Dispatchers and Center Managers can view previous logs of call data to inform future decisions.
Micro Interactions
Toggle on-and-off Modules Panel
Give operators more freedom and customizability with their interface to display necessary situational data at different parts of the workflow.
Pathfinding Interactions
Operators can monitor the path first-responders will travel based on live google maps data. They can see the traffic between each waypoint, view status updates, and recommend a change in the route if necessary.
User Feedback
Keeping the operator-in-the-loop by incorporating their feedback during decision-making. This data will be used to inform and adapt future AI recommendations.
Drag and Drop
To enforce customization and modularity, all panels are draggable and can be turned on and off depending on need. This increases situational awareness and enables more autonomy in deciding what info is most important.
Design System and Components
The design system is aligned with NATO Iconography and Color Guidelines. Widget components are modulated to fit multiple use cases for different emergency scenarios.
more to come soon...