GreenLens

Inspiration Corporate greenwashing is rampant as companies make bold reforestation pledges at press conferences and in ESG reports, but there's rarely an independent way to check if the forests actually recovered. We wanted to build a tool that cuts through the PR and puts satellite data behind the claims.

What it does GreenLens lets you type in any company name and instantly see an independent assessment of their reforestation or sustainability pledges. It automatically finds the company's most credible public commitment, geocodes the location, pulls real satellite forest-loss data from Global Forest Watch, and produces a truth score based on whether deforestation in that region actually slowed after the pledge was made. It also shows a before/after satellite map with a forest-loss heatmap overlay so you can see the change visually.

How we built it The backend is a FastAPI app in Python. When a company is searched, we use OpenAI's web search to find and rank candidate sources, official sustainability reports, ESG pages, and third-party articles, scoring each one on location precision, data completeness, and source credibility. The best source's location is geocoded with Nominatim, then we query the Global Forest Watch API for annual tree cover loss data in that region. A scoring algorithm compares average annual forest loss before and after the pledge period to produce a truth score. The frontend is React with Tailwind, using Leaflet for the satellite maps, MODIS Terra imagery for the historical "before" view, and ESRI World Imagery with a GFW Hansen loss overlay for the "after" view.

Challenges we ran into Getting consistent, structured data out of web search was harder than expected. Company pledges are described in wildly different ways, units (trees vs. acres vs. hectares), and time horizons. Many companies list global or multi-region commitments with no single mappable location, which breaks geocoding. We also had to be careful not to overstate what the satellite data actually proves: a drop in forest loss in a region correlates with a pledge but doesn't confirm causation. Natural disasters, droughts, and policy changes all affect the same signal.

Accomplishments that we're proud of Building a fully automated pipeline that goes from a company name to a sourced, satellite-backed assessment with no human-in-the-loop steps. The source ranking system, which scores and selects from a dozen candidate articles based on location precision, verifiability, and time window, ended up being surprisingly robust.

What we learned Satellite data is powerful but noisy. Forest loss measurements reflect many overlapping causes, and framing results honestly (correlation vs. causation) matters as much as the technical accuracy. We also learned that OpenAI's web search is great for finding sources but needs careful prompt engineering to extract structured, consistent data reliably.

What's next for GreenLens Year-filtered satellite imagery (via Sentinel Hub) so the before/after maps reflect the actual pledge period rather than a static snapshot. A company database so results are cached and comparable across organizations. Carbon credit verification not just reforestation pledges, but offset purchases like the Microsoft/Chestnut Carbon deal. And a shareable report card that journalists or researchers can link to directly.

Built With

• openai
• python
• react
• tailwind
• typescript