EcoSim

Dashboard home page
Graph simulator page

Inspiration

We wanted to leverage the power of Generative AI to improve the ecosystem. We focused on the California ecosystem, which is home to many of us.

What it does

An imagination-bending, chat-based ecosystem simulator game that challenges you to make a postivie impact on the California ecosystem. You'll interact with our LLM-powered chatbot, making crucial decisions and immediately witnessing how your actions influence the ecosystem’s delicate balance. From managing wildfire risks to balancing urban development, your choices determine the fate of California’s biodiversity.

How we built it

We combined the predictive power of a Bayesian network model with the conversational abilities of a finetuned LLM to create a truly immersive simulation experience. The Bayesian network explores how factors like climate conditions, urban expansion, wildfires, and more affect the health of the ecosystem, measured by indicators such as biodiversity, habitat connectivity, and species survival rates.

Challenges we ran into

    As one of the most ecologically diverse states, California’s datasets were multi-dimensional and broad in scope. This made data collection and preprocessing a challenge.
    From climate variables to species distribution and land-use patterns, aligning these datasets for meaningful predictions took meticulous effort before we could even start training the model.

Accomplishments that we're proud of

🌟 Building a fully interactive AI-powered simulator where users can see the impact of their decisions instantly. 🌿 Designing a Bayesian network that accurately models complex ecological relationships. 💬 Finetuning an LLM to serve as a fun, engaging host that makes learning about the ecosystem exciting. 🔄 Creating a game that is both educational and entertaining, sparking awareness about real-world ecological challenges.

What we learned

📚 Data matters: The importance of high-quality, well-aligned data in building reliable predictive models. 🔗 Complex systems need flexible models: How Bayesian networks are perfect for capturing uncertainty and causal relationships in dynamic environments. 💡 Engagement is key: The value of interactive experiences in making environmental education accessible and fun. 🧪 Iterate, iterate, iterate: The art of refining ML models through continuous feedback and testing.