π¦ Inspiration
It started with a craving. During the hackathon, we took a walk down Nassau Street to get some froyo and ended up talking with Brooke, Martina and Efren, longtime employees at Fruity Yogurt. They told us the toppings, the fruit and the boba pearls at their shop get thrown away the most, since they go bad within hours. No one had ever measured the waste, and making the usual amount just felt safest.
Wasted food also wastes the water, energy, labor and money that went into it, and it gives off methane in landfills. So we came back and built a dashboard just for their shop, using their own data, to help them make just enough and reduce the waste. π±
π± What it does
Forecast
- Predicts customers for the next 14 days
- Uses the weather, Princeton events, school breaks and HackPrinceton
- Learns from the shop's own numbers, logged by employees every night
- Gives each day a color-coded label, from "Very busy" to "Very slow"
Prep plan
- Tells the team when to start each boba batch and how big to make it
- Says how many fruit containers to cut and froyo machines to run
- Suggests what to add if the day gets busier than expected
- Estimates how much boba could be thrown out compared with the usual routine
Tools for staff
- Boba clock: a counter tablet in English and Spanish, since staff speak both
- Topping check: staff snap a photo of the topping bar and confirm what it shows
- Text assistant: text βprepβ to get tomorrowβs plan from anywhere
- Closing log: a 30-second check-in at close to record what was thrown out
π οΈ How we built it
We kept the forecast a simple, explainable product of factors:
$$\text{customers} = \text{typical day} \times \text{weather} \times \text{campus} \times \text{events} \times \text{correction}$$
- Reads 4 calendar sources every 30 minutes: iCal feeds parsed in code, web pages read by Claude
- Simulates boba stock hour by hour, with pearls expiring after a shop-set window (4 hours by default)
- Plans each batch from when stock runs out, minus the 45-minute cook time
- Moves the forecast 20% of the way toward the real count after each closing log
- Saves offline in the browser and merges every device's changes without overwriting
- Keeps a topping photo only if the check passes and no one is in frame, and deletes it after 14 days
- Marks every placeholder number "assumed" and every demo number "sample"
Built with
- Photon: iMessage texting assistant ("prep", "events", "help")
- Anthropic API (Claude): reads messy calendar pages and topping photos
- Computer vision: estimates how full each topping pan is, confirmed by staff
- Open-Meteo: live weather forecast
- BestTime.app (optional): hourly foot-traffic shape
- Node.js, JavaScript, HTML/CSS: server, API functions and dashboard
- libSQL / Turso: shared settings, events, boba batches and photos
- Vercel: hosting setup, with serverless functions and a daily cleanup job
π Accomplishments
- A working loop from forecast to prep plan to nightly learning
- Texting and tablet tools for people who won't open an app
- A forecast the shop can understand and trust
- A photo check that read 23 to 24 of 24 levels correctly on synthetic test images (not yet the real bar)
- 100 passing tests
π What we learned about forward deployed engineering
- Put the client's priorities first: we came to forecast customers, and the shop pointed us to the waste, so we followed them
- Show up and listen: our best ideas came from a conversation, not a dataset
- Meet people where they work: the help arrives right when someone decides how much to make
- Build for the person at the counter, because a tool only helps if people actually use it
- Be honest about what you know and what you're still guessing, because trust is what makes a solution stick
π What's next
- Fruity Yogurt can start using it now: add their real numbers, log each night, and let the forecast learn from their shop
- Measure waste before and after, so we can show their real savings
π€ Forward Deployed Challenge (Peri)
- Business: Fruity Yogurt, 166 Nassau St, Princeton
- People we consulted: Brooke, Martina and Efren, longtime employees at Fruity Yogurt
Built With
- anthropic
- claude
- computer-vision
- css3
- html5
- imessage
- javascript
- libsql
- node.js
- open-meteo
- photon
- sqlite
- svg
- turso
- vercel