EveryCent

🌿 Overview

EveryCent is an AI-powered grocery price forecasting app built for low-income families and SNAP recipients. It uses 25 years of real USDA retail price data and Google’s TimesFM 2.0 foundation model to predict whether essential grocery prices will rise or fall, turning complex forecasting into one simple decision: buy now or wait?

💡 Inspiration

Over 40 million Americans rely on SNAP benefits, and millions more live on tight grocery budgets. When staple food prices spike unexpectedly, families run out of money before they run out of month. Retailers already use sophisticated price forecasting to decide when to stock up, so we wanted to give that same intelligence to the families who need it most.

🛒 What it does

EveryCent forecasts grocery prices using real USDA retail price data and Google’s TimesFM 2.0 AI model. Users enter their monthly grocery budget and SNAP reload date, and the app tells them:

Buy Now: prices are rising, so stock up today
Wait: a price drop is coming, so hold off
Stable: no significant change is expected

Every recommendation includes an AI-generated explanation powered by Claude searching the web in real time. A personalized shopping list shows optimal buy dates for each item so families can plan their entire month in one view.

🛠️ How we built it

Data pipeline: 25 years of real USDA retail price data for eggs, milk, ground beef, bacon, chicken, potatoes, and bread. Prices are updated daily using Claude with web search to fetch the latest BLS and USDA figures automatically.
Forecasting: Google’s TimesFM 2.0 foundation model runs inside BigQuery ML through a single SQL query. No model training is required. We feed historical price data and get 14-day forecasts with 95% confidence intervals.
AI market context: Claude generates live alerts explaining why prices are moving, such as bird flu affecting eggs, drought conditions affecting beef, or seasonal patterns affecting bacon.
Frontend: React, TypeScript, Vite, Tailwind CSS, and Recharts for forecast visualizations.
Backend: Express.js server connecting to Google BigQuery for live price and forecast data.
Model evaluation: TimesFM achieves an $R^2$ of $0.91$ compared to $0.76$ for a moving average baseline, and an MAE of $\$0.06$ compared to $\$0.14$.

🧱 Challenges we ran into

BigQuery ML’s AI.FORECAST requires perfectly continuous time series data. A single missing month caused the entire forecast to fail after loading 150 rows of data.
Getting real USDA data into BigQuery required cleaning across multiple file formats and date conventions.
BigQuery ML does not yet support external covariates through SQL, which limited our ability to use corn prices as a leading indicator for meat and egg prices.
Coordinating parallel development across a 16-hour sprint caused repeated merge conflicts that required careful resolution.

🏆 Accomplishments that we’re proud of

Built a real end-to-end AI forecasting pipeline using actual government price data, not mock data.
Ran TimesFM on up to 25 years of real price history for some items.
Created a daily price updater that autonomously fetches current retail prices using Claude with web search and reruns the forecast.
Built a SNAP Cycle feature that aligns price forecasts with EBT reload dates, which is genuinely useful for millions of Americans.

📚 What we learned

Time series forecasting requires clean, continuous data. One gap can break everything.
EDA revealed that eggs and ground beef have dramatically higher price volatility, which directly shaped our modeling decisions.
The most impactful product features are the simplest to explain. “Buy now or wait?” resonates immediately with anyone who has worried about their grocery budget.

🚀 What's next for EveryCent

Expand to 500+ grocery items with automated daily price ingestion.
Add SMS alerts when a WAIT item hits its predicted low price.
Add store-specific pricing so users can find the cheapest nearby option.
Partner with food banks and SNAP outreach programs.
Integrate corn and soybean futures as leading indicators for meat and dairy forecasts.