Geoflare

Overview

GeoFlare is a collaborative ecosystem empowering real-time decision-making during live wildfires. Inspired by strides made in military defense startups like Palantir and Anduril, we developed a comprehensive, action-based platform to allow firefighters to focus on saving lives, not planning strategy.

The platform leverages geospatial computer vision and artificial intelligence to rapidly process environmental data and provide immediate, strategic actions to tackle wildfires effectively.

The Problem

As climate change exacerbates wildfires into a global issue, response systems have failed to catch up—remaining slow and unreliable. In 2021, wildfires burned over 7 million acres of land with 4.5 million people at risk of fire-related property damages.

Government solutions have tried to address this by developing web apps tracking broad geospatial locations. However, these maps fail to convey case-by-case wildfire risks, leaving individuals uncertain about their safety. The unintuitive software limits use cases, making the barrier-to-use extremely high.

The issue with wildfires wasn't a lack of resources, but rather an over-reliance on outdated, unsustainable methods of measuring fire safety. Government solutions merely identify the problem space but fail to convey strategic actions to combat them.

How It Works

GeoFlare provides three core capabilities:

• Identification: Live risk assessment based on objects identified in satellite imagery. We leverage computer vision to place boundary boxes around critical elements such as bush clusters that could create unanticipated impediments to firefighters.
• Live Chat: Live strategy co-pilot chatbot allows you to interact with a chat interface that recommends strategies based on the threats identified in the computer vision detection.
• Strategy: Path optimization by searching through an environment, leveraging information collected from computer vision, finds the most optimal route to a specific house address. It also measures the "priority level" of each home based on the severity of their wildfire situation.

Technical Stack

Built by a diverse team consisting of a full-stack developer, front-end developer, product designer, and ML researcher:

• Frontend: React, Next.js, TypeScript, Tailwind CSS, and SASS for the landing page
• Backend: Node.js with Gemini and PyTorch integrations for the ML model and chatbot interface
• Machine Learning: Custom-trained YOLO model for accurate, threat-based object identification consistent with danger-related information
• Design: Full design-thinking framework from ideation to secondary market research, user flow diagramming, low-fidelity wireframes, and high-fidelity prototypes in Figma

Design Process

The design was developed through a comprehensive design-thinking framework:

• User Flows: Mapped how users naturally flow through the page to adapt our product to quick thought-processing mechanisms
• Lo-Fi Wireframes: Helped map out structural layout and enabled quick iterations across different idea concepts
• Branding: Futuristic, geospatial branding inspired by defense companies like Palantir and Anduril
• High Fidelity Prototypes: Created a landing page along with final prototyped versions of the dashboard web app

Challenges

Designing for geospatial UI, computer vision, and dynamic animation states was challenging. Creating a multi-dimensional dashboard that combines static UI with live transformations of real-time data required careful consideration of user experience and information hierarchy.

On the technical side, training custom datasets to accurately identify threat-based objects and integrating multiple AI systems (computer vision, chatbot, path optimization) into a cohesive platform presented significant engineering challenges.

Accomplishments

We successfully built a functioning MVP with a full conceptual dashboard and landing page, creating a multi-dimensional, action-based platform based on real-time strategy and updates. The system demonstrates how modern AI and computer vision can transform emergency response coordination.

What's Next

Firefighters encounter many different and unpredictable experiences. We plan to expand our technologies to include more datasets, accounting for any possible scenario a firefighter may encounter by incorporating resources and recommendations tailored to each suggested strategy.

Training additional datasets including unpredictable and rare scenarios will provide firefighters with strategic predictions for never-before-encountered situations, making the platform even more robust and valuable in real-world emergency response.