While building our AI-powered yield forecasting and climate-based credit scoring system, we faced multiple challenges, including CORS errors, JWT authentication failures, auth token issues, cookie handling errors, and model inconsistencies.

CORS Error
Since our frontend and backend were on different domains, API requests were blocked. We resolved this by configuring Flask-CORS with supports_credentials=True and setting proper headers.

JWT Authentication & Token Issues
Our tokens were expiring too quickly or being rejected. We fixed this by adjusting expiration times, implementing refresh tokens, and ensuring tokens were properly signed and verified.

Cookie Handling Errors
Cookies weren’t being stored due to SameSite restrictions. We resolved this by setting secure=True, httponly=True, and SameSite=None for proper authentication.

Model Challenges
Our yield prediction model initially struggled with inconsistent geospatial data, leading to inaccurate forecasts. To address this, we tried ensemble learning models like XGBoost and Random Forest, but they lacked the ability to capture seasonal trends effectively.

At last, we switched to Prophet by Meta, which is designed for time-series forecasting and handles seasonality, holidays, and trends efficiently. This improved accuracy and reliability, making our predictions more useful for farmers and lenders.