Oasis

An Intelligent Urban Development Tool.

Inspiration

Climate Change has caused city planners face a "silent killer": the Urban Heat Island effect. Unsustainable building practices continue to plague cities; roads break down, power lines melt, and data centers were not built with intense heat in mind. Over 550,000 people will die from Extreme Heat, and this number will continue to grow by 250,000 every year. And considering the projected $500 billion lost from infrastructure repair, this problem is too big to ignore.

But there is a solution - construction agencies, governments, and city planners can use heat-resistant material, plant more natural cover, or engineer with solvency in mind, saving lives and money.

So why don't they? They don't have the information to know where and what to build with.

That's where Oasis solves.

What it does

Oasis is an intelligent urban heat management platform meant to supplement other analysis. It allows planners to:

Visualize Risk: Explore a high-resolution heat map of 492 census tracts and 25,552 blocks in King County.
Predict Outcomes: Access "optimal development" indicators such as Heat Risk, Tree Cover and Demographic Information. Access development recommendations generated by XGBoost and TensorFlow Models.
Solutions: Outlines optimal "sustainable" ways to build based on geography strengths/weaknesses.
Consult the AI: A built-in assistant (Haiku 4.5) that understands the map context, allowing users to ask natural language questions about development.

How we built it

Three Layers:

Data Ingestion:
Interpreting Geography: King County Heat Dataset allows the model examines temperature in the morning, afternoon, and evening for each tract in King County. It also analyzes distance to water, impervious surfaces, and tree canopy cover.
Interpreting Health: King County Health Dataset to comprehend full impacts of Extreme Heating, we layer in analysis of health factors - cardiovascular disease rate, diabetes rate, and life expectancy - and social vulnerabilities - poverty, disability rate, and income factors.

This will inform two critical factors: heat intensity and health risk.

Machine Learning
We first combine everything into a composite score to aggregate heat intensity and health risk. This is useful for comparison.
We use XGBoost (decision tree) to train models to understand combinations of features that are predictive of our desired output. This produces heat intensity and health risk.
We then take these two outputs and run a TensorFlow model (small neural network) with all possible features to get a composite output of total risk.
We then check; if our outputs all indicate the same levels of risk, it is a strong indicator. Finally, we use a composite formula to map relevant domains: $30% temperature + 25% vegetation + 25% health + 20% social factors.$ The weights are determined through research on importance of drivers. Every score lands between 0 and 1, and the higher the score is, the higher risk it is for building development. These values are stored in our DuckDB database to be queried later.
Platform Development City Planners, Governments, Construction Companies, and any stakeholders can use our product for a: a. Map View - heatmaps show optimal building locations. b. Ranked List - sorts tracts by scores to find high priority neighborhoods. c. Recommendations - can use the chatbot to ask any questions about the data and outputs. d. Projections: glances 25 years into the future to walk stakeholders through best regions to build in.

Challenges we ran into

The Information Limit: We wanted our AI assistant to "see" every neighborhood the user was looking at. However, sending detailed data for 50+ areas at once was too much information for the AI to handle in one go.
The Fix: We created a "shorthand" code to shrink the data size. This lets the AI understand the entire map without hitting a text limit.
Software Clashes: We ran into a major roadblock where the latest versions of our AI tools didn't play well with the newest Mac computers, causing the app to crash constantly.
The Fix: We switched to a more stable version of our coding environment and changed how the app "wakes up" so it doesn't freeze while loading its brain.
Slow Loading Maps: Drawing 25,000 city blocks on a map is a massive task. Initially, it was too slow because the system had to translate the data into a format the map could understand one by one.
The Fix: We shifted that "translation" work directly into our database. Now, the map data is ready to use the instant it's pulled, making the dashboard feel snappy and responsive.

Accomplishments that we're proud of

Machine Learning Workflow We successfully built a system that translates complex climate science into simple risk scores. It takes massive datasets and turns them into a "weather report" for every single city block in King County, making it easy for anyone to understand which areas are in danger.
Sub-Second Simulations Usually, climate models take forever to "re-think" when you change a variable. We built an engine based on the latest 2024 heat research that gives planners instant feedback. If you "plant" trees on the map, you see the cooling effect immediately, not five minutes later.
Heatmap UI We designed the system to be incredibly lightweight. By keeping the heavy data processing behind the scenes, the map stays snappy and responsive, even when it’s juggling data for tens of thousands of city blocks at once.

What's next for Oasis

Live AI Insights We’re upgrading the chat so it responds instantly as you browse. As you click through different neighborhoods, the AI will give you a real-time, play-by-play analysis of the local heat risk without you having to ask.
The Neighborhood "Lasso" We’re adding a tool that lets users "lasso" or circle any custom area on the map. This will give planners an instant report on heat risk and the best ways to improve that specific patch of the city.
Beyond King County The system is built to scale. We plan to allow users to upload their own local data so Oasis can provide the same deep analysis for any city or county in the country.
New Building Profiles Heat doesn't just affect parks. We’re expanding our models to look at different types of buildings—like schools, hospitals, and low-income housing—to see how specific structures can be upgraded to handle rising temperatures.
The Budget Sandbox We’re building a financial tool to help cities put their money where it matters most. It will calculate exactly how a specific budget - like a million-dollar grant - should be spent to save the most lives and cool the most blocks.