Doorstep

Every customer deserves a response.

Inspiration

When a disaster strikes, emergency alerts are only effective if people can receive and respond to them.

Disasters affect entire communities, but their impact is not distributed equally. Older adults living alone, individuals with disabilities, tourists, international students, and displaced renters can face additional barriers to receiving assistance.

We wanted to approach disaster response from the perspective of an insurance provider. Instead of simply identifying where a disaster occurred, we asked: How can State Farm understand which customers are currently being affected, identify who needs attention first, and coordinate meaningful assistance?

That question inspired Doorstep.

What it does

Doorstep is an AI-integrated customer impact assessment and disaster response coordination platform designed to help insurance providers understand how disasters are affecting their customers in real time.

Doorstep combines multiple sources of information to assess current customer impact, generate explainable rankings, and help response coordinators determine where attention is needed most.

Our platform has three core capabilities:

1. Signal Fusion and Customer Impact Ranking

Instead of relying on a single indicator, Doorstep combines disaster intelligence, infrastructure disruptions, customer circumstances, and communication signals to assess the likelihood that a customer is currently affected.

Each customer receives an explainable rank based on available evidence. Coordinators can see exactly which signals contributed to that ranking, rather than relying on an unexplained number.

For example, a customer who has stopped responding while experiencing a neighborhood power disruption and has a documented power-dependent medical need may require a different response than someone experiencing an isolated communication failure.

2. Interactive Disaster Response Map

Doorstep visualizes customer impact geographically through an interactive map. Response coordinators can explore affected neighborhoods, identify households requiring attention, and access a dynamically organized priority queue.

3. Personalized Outreach and Assistance

Doorstep adapts its outreach to individual customer circumstances rather than sending everyone the same emergency alert.

Through Photon-powered messaging and voice calls, the platform can:

Send personalized emergency check-ins.
Provide multilingual communication.
Escalate unanswered check-ins through alternative communication channels.
Coordinate with authorized emergency contacts.
Classify incoming responses to identify customers who are safe or need assistance.

When a customer needs help, Doorstep can also connect them with relevant resources, including accessible shelters, evacuation assistance, and post-disaster insurance claim guidance.

How we built it

We built Doorstep using a full-stack architecture that connects customer information, disaster intelligence, AI-assisted processing, and real-time communication.

Frontend and Application

TanStack Start (React + Vite): Frontend and server API routes
Tailwind CSS and shadcn/ui: Responsive interface with a red-and-white liquid glass design.
Framer Motion: Interactive transitions and real-time dashboard animations.
MapLibre GL + OpenFreeMap: Interactive 2.5D map of Texas with 3D buildings, used to visualize affected households and neighborhoods.

Backend and Data

Supabase: Stores escalation events and response history.
Synthetic customer records: Simulated households, customer circumstances, and disaster scenarios to demonstrate the platform.

Communication and AI

Photon: Real text check-ins and reply handling
NVIDIA Nemotron: Classifies resident replies (including non-English) and generates case summaries.

Disaster Intelligence and Resources

Ting: Power disruption intelligence, demonstrated through simulated outage events where direct data access is unavailable.
FEMA: Public shelter information and accessibility data.

Our architecture separates AI-assisted decisions from the core escalation logic, allowing us to switch between AI providers while maintaining consistent application behavior.

Challenges we ran into

One of our biggest challenges was moving beyond a simple unanswered-message system.

An unanswered message does not necessarily mean someone needs emergency assistance. It could indicate a dead phone, a neighborhood power outage, or a communication network disruption.

We wanted Doorstep to distinguish between these situations instead of automatically treating every unanswered check-in as an emergency.

This led us to develop our Signal Fusion approach, which considers multiple sources of evidence when assessing customer impact.

Another challenge was designing a system that accommodates people with different needs. A generic emergency notification cannot account for mobility limitations, language barriers, transportation challenges, or power-dependent medical equipment.

We addressed this by incorporating customer preparedness profiles and adapting outreach and resource recommendations to individual circumstances.

We also had to work around data-access limitations. Some potentially valuable sources, including Ting do not provide unrestricted access to individual-level information. We designed our demonstration around synthetic customer profiles and simulated events while using publicly available disaster and shelter information where possible.

Accomplishments that we're proud of

Moving beyond traditional emergency alerts: We built a system focused on assessing current customer impact rather than simply distributing notifications.
Developing Signal Fusion: We combined multiple independent signals to generate more meaningful and explainable customer impact rankings.
Creating a unified response dashboard: We brought geographical visualization, customer profiles, and prioritized outreach into one interface.
Designing adaptive communication: We incorporated multiple communication channels and personalized outreach to account for customers who may not be able to respond to standard text alerts.
Connecting identification to assistance: We designed the platform to go beyond identifying affected customers by connecting their needs to relevant resources and recovery guidance.
Building with real-world insurance applications in mind: We designed Doorstep around the needs of insurance response coordinators and the customers they support.

What we learned

Throughout this project, we learned that effective disaster response requires more than collecting information or sending alerts. It requires understanding how different signals relate to one another and recognizing that missing information does not necessarily mean a customer is safe or unaffected.

We also learned the importance of designing AI systems that are explainable and appropriately limited. AI can help interpret unstructured customer responses and recommend actions, but the system should not treat uncertain information as confirmed facts or replace human judgment in emergency decisions.

Finally, we gained a better understanding of how insurance providers can use technology to support customers beyond the traditional claims process, particularly during the critical period immediately following a disaster.

What's next for DoorStep

We envision expanding Doorstep beyond immediate disaster response and into the entire customer recovery journey.

Our future plans include:

Deeper insurance integration: Connecting authorized State Farm customer and policy information to provide more context when assessing customer impact.
Improved signal integration: Incorporating additional verified infrastructure, weather, and disaster intelligence sources.
Advanced resource matching: Improving recommendations based on accessibility, transportation, medical, and housing needs.
Insurance claim guidance: Helping customers document property damage, understand claim requirements, and navigate the next steps after a disaster.
Continuous impact assessment: Updating customer rankings as new information becomes available and their circumstances change.

Our long-term vision is to make Doorstep a platform that helps insurance providers move from reacting to reported damage toward understanding customer impact and coordinating assistance when it matters most.

Built With

TanStack Start, React, TypeScript, Vite, Tailwind CSS, shadcn/ui, MapLibre GL, OpenFreeMap, Supabase, Photon (spectrum-ts), NVIDIA Nemotron, FEMA National Shelter System, simulated Ting-style outage data