Foresight

Screenshot of Foresight

Foresight

The disaster has not happened yet. We have already mapped it.

Inspiration

The January 2025 Pacific Palisades wildfire exposed a dangerous gap in emergency response: people often do not die from the initial disaster alone, but from the cascading infrastructure failures that follow.

A wildfire can damage power lines, trigger substation outages, disable traffic signals, block evacuation routes, and create secondary hazards like debris flows—all across agencies that traditionally operate in silos.

We asked:

What if emergency responders could see the full cascade before it happens, with AI recommendations they can actually trust?

That became Foresight.

What It Does

Foresight is a multi-agent emergency response coordination platform that models wildfire-driven infrastructure cascade failures in real time.

It helps emergency operators understand not just where a wildfire is, but what that wildfire will do next.

For our demo scenario, Foresight simulates the Pacific Palisades wildfire and predicts a cascading failure where:

Fire threatens transmission infrastructure
Power infrastructure fails
Traffic signals lose power
Evacuation routes become blocked
Secondary hazards, including debris flow, emerge

Instead of showing isolated incidents, Foresight provides a unified operational picture across agencies.

How We Built It

Foresight was built as a distributed multi-agent emergency intelligence system with a simulation dashboard, orchestration backend, deterministic validation engine, and AI coordination layer.

Frontend

Built with:

React
Vite
Leaflet
Interactive map simulation
Cascade replay timeline slider
Validator event feed
Agency command panels

The frontend visualizes wildfire spread, infrastructure failures, evacuation impact, and cross-agency recommendations in a real-time operations dashboard.

Backend

Built with:

FastAPI
Python
Server-Sent Events (SSE)
Structured JSON contracts
Timestep simulation orchestration

The backend manages simulation state, event propagation, agent execution, and communication with the frontend.

AI Layer

Our AI coordination system includes:

Hazard Agent
Cascade Agent
Secondary Hazard Agent
Coordinator Agent

Featherless AI API
IBM watsonx.ai Granite

Each agent specializes in a different operational domain, allowing parallel reasoning and coordinated emergency recommendations.

Physics Validation Layer

A major differentiator is deterministic validation.

Foresight validates AI outputs against physical reality before recommendations reach operators.

Validation models include:

Rothermel wildfire spread modeling
Deterministic infrastructure dependency logic
USGS debris-flow probability modeling

Debris flow risk is evaluated conceptually as:
P = f(slope, burn severity, rainfall intensity)

If an AI recommendation conflicts with known physical constraints:
Foresight rejects the output and forces a replan.

This prevents hallucinated emergency recommendations from reaching decision-makers.

Challenges We Ran Into

The biggest challenge was balancing ambitious real-world disaster modeling with hackathon execution speed.

Key challenges included:

Designing a believable multi-agent architecture under extreme time constraints
Integrating deterministic validation with probabilistic AI outputs
Synchronizing frontend simulation visuals with backend state changes
Modeling infrastructure dependencies clearly enough for judges to understand instantly
Building a dashboard that feels like a real emergency operations centre
Managing integration across parallel team workflows

One especially difficult challenge was explainability.
In emergency response: > "AI says so" is not acceptable.

Operators need reasoning they can trust—not black-box outputs.
Building trust became just as important as building intelligence.

Accomplishments That We're Proud Of

Physics-Validated AI Recommendations

Foresight checks AI-generated recommendations against deterministic physical models before surfacing them.
This dramatically improves reliability in high-stakes scenarios.

Cross-Agency Coordination Emergency response is often fragmented.

Foresight brings together:

Fire response
Utility operations
Traffic management
Evacuation planning into a single coordinated operational picture.

Clear Simulation Storytelling

The infrastructure cascade is visually understandable within seconds:
Wildfire → Power Failure → Traffic Failure → Evacuation Blockage → Secondary Hazard

That clarity makes the platform both technically compelling and immediately understandable.

Multi-Agent Architecture

We successfully built a distributed AI coordination system where specialized agents collaborate instead of relying on one monolithic model.

What We Learned

This project taught us that emergency AI systems need more than intelligence.
They need trust.

We learned that:

LLMs are powerful for coordination, but require guardrails
Deterministic validation dramatically improves reliability
Visualization is just as important as backend sophistication
Multi-agent systems are effective for decomposing complex emergency workflows
Disaster response requires interoperability across domains

Most importantly:

Impactful AI is not about replacing experts. It is about augmenting decision-making under pressure.

What's Next for Foresight

Foresight currently focuses on wildfire infrastructure cascade response, but the architecture is extensible.
Next steps include:

Real-time live incident ingestion
Flood response modeling
Hurricane infrastructure cascade simulation
Earthquake emergency coordination
GIS infrastructure integrations
Emergency notification system integration
Predictive evacuation route optimization
More advanced infrastructure dependency modeling
Incident commander collaboration tools
Multi-region deployment support

Our long-term vision is to make Foresight a trusted operating system for disaster coordination: Transparent, validated, AI-assisted emergency decision-making when every minute matters.