WasteWise AI

Smarter meals. Less waste. A greener future. WasteWise AI helps our school's cafeteria to prepare and buy the right amount of food to reduce waste, cost, and environmental impact.

Comment

Inspiration

Business optimization — connecting DECISIONS to IMPACT.

What it does

WasteWise AI is a system designed to help cafeteria managers in our school reduce food waste by improving how much is bought and prepared each day. It starts with historical BRMS data — tracking what was prepared versus what was actually served — along with a planned menu for upcoming days. A Python-based prediction model uses this data to forecast demand for specific items (like apples and Caesar salad). Those forecasts feed into optimization engines that recommend purchase and preparation quantities, factoring in things like leftover inventory from previous days (carryover), shelf life, and safety stock, so the system doesn't simply re-order based on demand alone. The results — a buy plan, a prep plan, and a waste dashboard — give cafeteria managers clear, actionable recommendations. Throughout the process, a generative AI layer (ChatGPT) stays connected to the BRMS data so managers can ask plain-language questions about any forecast, plan, or trend at any time. Final decisions always remain with the human manager, who can review, adjust, or override the AI's recommendations before anything is purchased or prepared.

How we built it

  1. Asked questions to cafeteria staff at Bridgewater-Raritan Middle School, researched topics(environmental impact), and collected the full menu from the school website to generate a reality-based synthetic dataset (Prepared − Served = Waste).
  2. Analyzed this data and created a report on total waste and environmental impact.
  3. Created a forecast engine in Python to predict demand for the next few weeks.
  4. Created a buy optimization framework that includes carryovers.
  5. Built an Apple Optimizer and Buy Sheet.
  6. Made a Chicken Caesar Salad Optimizer.
  7. Troubleshot opposite results on the optimizer.
  8. Built a better Chicken Caesar Salad Optimizer and Prepare Plan.
  9. Connected everything.
  10. Made Gen AI (ChatGPT) analyze it.
  11. Built a final saved waste dashboard. ## Challenges we ran into During testing, we found that the RMSE and MAE values on the Chicken Caesar Salad Optimizer were miscalibrated, causing it to produce more waste than the original process — the opposite of its intended goal. After investigating, we discovered the two error metrics were unbalanced, pushing the optimizer toward unstable recommendations. We re-tuned both variables, and the optimizer not only returned to normal performance but outperformed our original expectations, generating prep recommendations that reduced waste even more effectively than projected! This troubleshooting step taught us how directly forecast error impacts optimization outcomes!

Accomplishments that we're proud of

Apple waste per serving dropped from 24% to 11%, meaning 13% reduced waste, and overall 55% decrease. Chicken Caesar Salad Waste per serving dropped from 17% to 9% meaning 9% waste reduction and overall 47% decrease.

What we learned

How to apply predictive, prescriptive, Gen AI to solve a real life problem.

What's next for Waste Wise AI

All Item WasteWise In-built attendance and weather forecasting. Waste audit(checking plates) integration Safety stock forecasting and optimization
Self menu creation Health-based optimization Full pipeline automation

Built With

Share this project:

Updates