SodaPy.com

What it does

I fine-tuned Gemini with a government dataset. dataset was processed: feature engineering, tests, etc. I combined my fine-tuned Gemini with ARIMA and XGboost models to create an Ensemble of 3 models. This brought down my MAE to about 6. At first I was getting predictions from all US states but that was a lot of API data requests-which slowed down the website. But fine-tuned Gemini did very well predicting soda consumption for all states alone. The fine-Gemini did very well with predictions but with a very high MAE. So ARIMA and XGBOOST was mostly used to bring down the MAE significantly, but the fine-tuned Gemini alone was already producing great forecasting soda consumption results when compared the original government dataset. So the real power in soda consumption predictions comes from fine-tuned Gemini. Texas is where I had my surgery to donate my kidney to my father because of his Kidney Failure. So its nice to focus on Texas future soda consumption to tackle the kidney issues here. Now that we are aware of soda consumption, we can ask people to anonymously/voluntarily donate a urine test which is the best tool to measure Stage 1 Kidney Failure. Stage 5 kidney failure is already too late, and people need surgery at Stage 5. So with soda consumption predictions, we can focus on Stage 1, prevention approach. Kill Kidney Failure at its root. I used Vertex AI an GCP Cloud Run/Load Balancer. I used Chrome dev tools to check latency API issues.