Finding Forester

Your AI-powered environmental impact assessment tool

Comment

Inspiration

Around 15 billion trees are cut down every year globally, driven primarily by agriculture, logging, and infrastructure expansion. While around 5 billion trees are planted or regenerated, the planet still experiences a net loss of approximately 10 billion trees annually. Finding Forester (influenced by but not related to the screenplay Finding Forrester) was born from the idea that anyone should be able to point at a map and instantly visualize the true "cost" of losing a specific patch of land. By identifying the specific inhabitants and ecological services of every forest, we aim to provide a clear, data-driven look at the real pros and cons of land use, allowing users to move from awareness to actionable conservation or businesses to understand the true scale of their operations.

What it does

Finding Forester is a geospatial AI assessment tool. Users interact with a global map to draw polygons around specific forested regions. The app then:

  • Analyzes the Terrain: Identifies hectares, coordinates, and proximity to urban centers.
  • Generates Ecological Deep-Dives: Using Gemini 2.5 Flash, it produces reports on biodiversity loss, soil degradation, and specific threats to native flora and fauna.
  • Predicts the Future: Offers decade to century long climate outlooks based on regional data.
  • Prescribes Recovery: Provides localized restoration plans, including specific tree species for replanting and estimated conservation costs.

How we built it

The application was built using Next.js and Tailwind CSS for a responsive, modern interface. We used map integration using Leaflet and OpenStreetMap which allows for custom polygon drawing and coordinate extraction. These coordinates were passed to the Google Generative AI SDK, where we engineered specific prompts to return structured, professional-grade ecological reports.

Challenges we ran into

  • Geospatial Logic: Converting a hand-drawn map polygon into a format the AI could actually understand was a major hurdle. We had to carefully process coordinate arrays and calculate hectares before sending anything to the model.
  • AI Hallucination: Early tests had the AI suggesting generic or "made-up" plants as well as overexaggerating on the effects of deforestations when managing smaller forests. Furthermore, we had to fix the issue were it would estimate the number of trees when provided with an coordinates in an ocean. To fix this, we implemented strict system instructions that forced the model to cross-reference coordinate and regional ecological data to ensure the data it was providing was as accurate as possible.
  • Cost Realism: We had to fine-tune our prompts to ensure the AI provided realistic conservation budgets rather than exaggerated solutions.

Accomplishments that we're proud of

  • Map-to-AI: We successfully built a system that takes raw map coordinates allows Gemini API to produce informed, data-driven reports in seconds.
  • Data Retention: We implemented a persistence system using localStorage and Cookies so users don't lose their ecological assessments when reloading the page.
  • UI/UX: We created a simple yet elegant UI that displays proper information while being structured and easy on the eyes.

What we learned

Working on this project at HackPSU 2026 taught us that AI can be used for so much more besides a chat bot or homework help. It has potential to solve some of the world's most pressing issues given enough time and data. For us specifically, we discovered how to use it as a powerful engine for analyzing maps and environmental data.

We also got important experience with mapping APIs and learned how to take things like raw coordinates from a user's mouse click and turn them into exact data to make accurate AI-reports.

What's next for Finding Forester

  • Real Satellite Overlay: Integrating live deforestation data (like Global Forest Watch API) to show "Change Over Time."
  • Oceanic Expansion: By integrating bathymetry data (ocean depth) and sea-surface temperature layers, we can allow users to select marine protected areas (MPAs), coral reefs, or coastal mangroves.
  • Fine-Tuning: We plan to move beyond general LLM responses by training our models on specific ecological datasets and peer-reviewed conservation papers. This will ensure that every recovery plan is backed by the latest environmental science and that we are getting more accurate estimates about the true environmental impact of deforestation in a given area.
  • PDF Export: Allowing users to download the generated reports as official-looking PDF documents for school or local government presentations.
Share this project:

Updates