Hero Page
Barrier Map
AI Engine Selection and Analysis
Responsible use of AI
Data Sources

Inspiration

Greenlight was inspired by a simple question: how can schools know which sustainability improvements will actually make a difference? We wanted to build something that helps schools make smarter decisions instead of guessing, especially when resources are limited and every investment matters.

What it does

Greenlight is a decision-support platform built for Bay Area high schools.

Users select a school on the map, and Greenlight analyzes the surrounding area using a deterministic scoring system based on accessibility, safety, environmental importance, equity, and feasibility.

From there, schools can simulate improvements like installing bike racks, adding protected crosswalks, or improving recycling infrastructure. The platform recalculates projected barrier scores and expected impacts instantly, allowing users to compare different ideas before investing time or money.

Rather than generating recommendations from scratch, Greenlight uses AI to explain the results, summarize the reasoning behind each recommendation, and present the information in language that's easy to understand.

How we built it

We built Greenlight as a modular, data-driven system with separate layers for scoring, processing, and explanation. This made the project easier to manage, easier to debug, and easier to improve later.

The core of the project is a deterministic scoring engine. Each location is evaluated across factors like accessibility, safety, environmental importance, equity, and feasibility. Those inputs are combined into a final score, which keeps the output consistent and avoids relying on AI to make the actual decisions.

We then built a data pipeline using real OpenStreetMap data through the Overpass API. That gave us live geographic and infrastructure data to evaluate locations more realistically. The structure is flexible, so other public datasets can be added later without changing the main logic.

The AI layer is only used to explain the results in plain language. It takes precomputed scores and turns them into clear summaries, which helps reduce hallucinations and keeps the system transparent. On the frontend, we focused on presenting the results in a simple way so users can quickly understand the ranking and the reasoning behind it.

Challenges we ran into

One of our biggest challenges was deciding where AI actually added value.

At first, we considered letting the model generate recommendations directly. The more we tested, the more we realized that wasn't the right approach for a project involving decision-making. Schools should be able to trust where every number comes from.

That led us to redesign our architecture so that every calculation is deterministic and the AI is responsible only for explaining the results.

Another challenge was balancing technical depth with usability. We wanted Greenlight to perform meaningful analysis without overwhelming users, so we spent a lot of time simplifying the interface and presenting complex information through maps, visualizations, and simulations instead of large blocks of text.

What we learned

We learned that AI is most useful when it supports the system instead of controlling it. By keeping the scoring deterministic and using AI only for explanation, we made the project more reliable and easier to trust.

We also learned how to work with real geospatial data and turn raw location data into something useful. Using OpenStreetMap and Overpass taught us how important data cleaning, filtering, and feature selection are when building a practical product.

Another significant lesson was that scope matters. Focusing on Bay Area high schools helped us build something realistic and meaningful instead of trying to solve every use case at once.

What's next

Greenlight is currently focused on Bay Area high schools, but the framework could be expanded to support additional schools, districts, and communities globally.

Going forward, we want to replace our current scoring inputs with more complete real-world datasets and validate the model against feedback from schools and local agencies. We also want to expand the platform beyond a single use case by adding richer geospatial signals, better transit data, and more detailed infrastructure features. The goal is to make Greenlight more accurate, more scalable, and more useful for real sustainability planning.