TerraLend

climate scientist panel p 1
climate scientist panel p 2
farmer panel p 2
farmer panel p1
loan officer panel, small part
climate scientist
climate scientist panel
loan officer panel p 1

Inspiration

During our brainstorming process, we were drawn to agriculture through an unlikely connection. While using an agricultural product during our ideation session, one of our team members was reminded of their time volunteering at a farm, and the discussion eventually turned to how it would be extremely beneficial if we could find a way to alleviate the stress that weather fluctuation causes on a farm. This led us to another question: if a farmer's entire livelihood is determined by what the climate does, why does the financial system that funds them completely ignore it? Loan officers today use credit scores, historical yield tables, and regional averages, which are static tools that don't fully reflect life on a farm, which leads to farmers defaulting on loans in case of a climatic emergency. TerraLend was built to resolve this question, making loans dynamic to the land. This gives farmers transparency and fairer rates, while giving lenders real-time intelligence to predict risk before it becomes a default.

What it does

TerraLend uses real-time data from NOAA CDO and Open-Meteo to provide live climate data such as NDVI, rainfall, soil moisture, and temperature. TerraLend utilizes machine learning and AI to provide dynamic interest rates through stress scores and recommendations for farmers/financial lenders regarding loans, risk portfolios, interest rates, etc.

How we built it

Back End: Python/FastAPI, SQLite to store both climate and loan data, boto3 to connect to AWS Bedrock.
Front End: React/TypeScript, Tailwind CSS, Leaflet to map the data, Recharts to view the trend data.
Data Sources: NOAA CDO, Open-Meteo for collecting real-time climate information.
AI: An XGBoost ensemble model is used to predict crop stress, and Claude is used to create narrative descriptions of how the loan terms will be adjusted.

Challenges we ran into

Resolution of Data: Our MVP uses regional climate data; we want to collect farm-level climate data. We have confidence bands to help mitigate this and a roadmap for developing parcel-level precision.
Noise in Signals: Drops in NDVI can give false signals. We cross-reference signals, use trends, and take into account crop cycles.
Fairness in Dynamic Pricing: Changes in rates could unfairly punish farmers during times of crisis. We implemented rate smoothing and hardship modes.
Regulatory Concerns: There needs to be transparency regarding automated predictions. We have audit logs and approval processes that involve humans to ensure that the system is working fairly.

Accomplishments that we're proud of

We successfully connected real-time climate data with predictive loan terms.
We developed an end-to-end system that is powered by AI, transparent, and easy to use.
We created narratives that are understandable by both farmers and lenders to describe why certain loan terms were chosen.

What we learned

Developing a system that combines AI, climate science, and finance requires serious thought regarding the ethics, fairness, and clarity of the decision-making process. 
A machine learning model such as XGBoost can have a large impact when combined with good explanations and some form of human oversight.
To develop a complex system within the timeframe of a hackathon, teamwork and an assigned role were absolutely necessary.

What's next for TerraLend

We plan to add more Agentic AI capabilities so that agents can take actions for farmers and financial lenders without them even having to worry about the loan process and risk portfolio maintenance.
We plan to move from regional climate data to parcel-level climate data using satellite and IoT data. - - - We also plan to include additional factors of risk, such as irrigation availability, commodity price, and insurance coverage. 
Finally, we plan to explore expanding into other climate-sensitive lending markets while maintaining a commitment to fairness and transparency.
We plan to add adaptive loan contracts that automatically flex repayment schedules when climate stress crosses a defined threshold -> this will be integrated with the agentic AI's new capabilities
We plan to add global region support covering more farms outside of just the US