DECISION SUPPORT TOOL
Designing a platform for AI Agents to offload 911 calls for emergency responders
I worked with dispatchers to design a platform to handoff non-emergency calls between human operator and AI agent. Won the largest AI hackathon in the world and received $68,000 in grants from Skydeck, Intel, and OpenAI.
Role
- Co-Founder
- Product Design
Duration
10 months
Team
- 3 Engineers
- 1 Designer
Results
- Grand Prize, world's largest AI Hackathon
- $68,000 investment
- Berkeley Skydeck Pad-13
winning the largest AI hackathon in the world
Dispatch AI started from a spontaneous trip with 3 friends at the Berkeley AI Hackathon. We won the Grand Prize and $18,000 in credits from Intel and OpenAI. Post-Hackathon, we decided to take this project further and launched a venture with Berkley Skydeck with $50,000 in investment. I worked as the solo designer.
shipped
I designed V1 of Dispatch in under 36 hours at a hackathon. I built out the entire platform afterwards.
- + fully equipped design system
- + full emergency response platform: incident monitoring, call handling, data analysis, and logging
- + database management system: built new data processing pipeline for real-time call handling
- + event forecasting model: custom-trained model on historic call data
final design
the operational loop
The system has two separate agents: 1) the human agent (dispatcher) takes emergency calls. 2) the AI agent handles non-emergency calls.
I designed Dispatch on a continuous operational loop that decides which agent should “take over” at each decision point (see model design for more details). Each session data is logged to inform future decisions.
01 live incident monitoring
Operationalize transcripts by supporting live language translation and dynamic script recommendations. Trained on 1000s of emergency call data.
02 dispatching units
Select and dispatch units on-platform. Select from pre-designed emergency procedures to reduce on-site decision-making.
03 Resolving alerts
Currently, calls themselves act as decentralized alerts. We dedicated a separate tab for alerts, alongside details and checklists, into a single pane-of-glass.
04 Customizable modules
Currently, dispatchers navigate across multiple static interfaces. We allow them to toggle on-off modules based on the data they need to surface during the current stage of operations.
05 Pathfinding
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 changes.
06 Historic call data
Previously, dispatch centers store old call scripts in thousand-page albums. We created a first-class solution that stores all previous call in a central database. Operators can change security settings to configure visibility.
07 Forecast future call volumes
Provide dispatchers with projected call volumes in the following weeks to help anticipate spikes in calls. Based on historic call volumes, which dispatch centers already collect.
the problem
The grim reality of dispatch centers: 82% are understaffed, with the average wait time of Oakland PD being 62 seconds. 90% of these calls are non-emergency, creating a system where dispatchers are overwhelmed with routine calls while critical emergencies wait.
This constant pressure takes a severe toll on dispatchers' mental health and operational efficiency.
"We get a lot of non-emergency calls that distract us from critical calls."— LAPD Deputy Chief
"I feel chronically anxious and stressed, which has impacted my mental health"— Dispatcher
at ground 0
We had the opportunity to speak directly with LAPD staff and observed common trends across dispatch centers. Through our research, we identified three key pain points that affect dispatchers daily.
user archetype
The archetypical user sits on two ends of the spectrum:
- Veteran dispatchers with decades of tenure
- New-hires with no habitual reflexes toward dispatch ops
However, both parties are overwhelmed by the amount of data they need to consume on a daily basis. This surfaces most when reading off manual instruction scripts during emergencies.
current workflow
There are 3 primary decision points in a call. Currently, this work is entirely manual - dispatchers type in tiny textboxes and read off scripts. The process is inefficient and error-prone.
the root node
Poorly-documented calls is the root cause of distress. The symptoms of this include:
- slower response times
- operator stress from repetitive calls
- manual and in-malleable response procedures
The opportunity, therefore, is offloading documentation to a more accurate and efficient agent to free up time for dispatchers to focus on critical calls
solution smithing
big picture: communication flow
The communication workflow flows through a chain-of-command system from the caller to the dispatcher, incident commander, unit commander, and finally the first responders. All of this intel must also overcome physical barriers: sensor systems, gps, wifi signals, and bluetooth.
This helped me better understand where each window and task can fit inside of the comms workflow.
appendix
design system
custom-built branding and component library. Referenced from NATO symbology/color guidelines.
