Gemma 4 Good Hackathon · Retail + Food Waste

AI production copilot for retail chains.

BakerySense combines a quantile demand model with a Gemma 4 agent to tell merchants exactly how much to produce and restock each day, with plain-language explanations grounded in real drivers. Multi-tenant, multi-branch, runs on Cloudflare.

Sign inCreate a tenant
Forecast accuracy uplift
−27% WAPE
LightGBM q=0.5 vs seasonal-naive
SKUs beaten baseline
19 / 20
French Bakery Kaggle dataset
JS↔Python parity
700 / 700
within 1e-4 absolute
End-to-end latency
~5–15s
Gemma 4 tool-calling turn

Sample exchange

Manager: How many TRADITIONAL BAGUETTE should we bake tomorrow at Quito Centro?
BakerySense: Bake 135. The model forecasts q=0.7 of 135 units, driven by lag_7=+46 (last Thursday was strong) and rolling_mean_7=+29.