DonateSmart

Inspiration

Goodwill sorts millions of donations manually with no visibility into what's coming. Donors don't know what's actually needed — so staff waste hours sorting items that were never acceptable. We asked one question: what if sorting happened before the donation left the donor's house?

What it does

DonateSmart lets donors photograph and pre-screen items before arriving at drop-off. Gemini 2.5 Flash verifies the item, assesses condition, generates a description, and assigns a price range — then issues a QR code the donor prints and attaches. Staff see every incoming donation on a live dashboard before it arrives. One scan at drop-off confirms the item. Door to floor, faster.

How we built it

React frontend, FastAPI backend, SQLite. Gemini 2.5 Flash for image classification, condition assessment, and auto-pricing. ElevenLabs for voice-guided donation — filling the form hands-free in real time. QR codes link each item to its full record. Donor loyalty tracking, anonymous donation support, and a staff approval workflow — all wired together in 48 hours.

Challenges we ran into

Prompt engineering Gemini to reliably distinguish poor lighting from actual damage without over-rejecting good donations. Syncing ElevenLabs voice state with React form state in real time. And honestly — the hardest challenge was scope discipline. Cutting genuinely useful features to stay focused on the core problem, repeatedly, under time pressure.

Accomplishments that we're proud of

A working end-to-end system — not a mockup. Donor submits, Gemini analyzes, QR generated, staff dashboard updates live, item tracked to inventory. The rejection flow specifically — when a photo doesn't match, DonateSmart tells the donor what it sees and redirects them to something Goodwill actually accepts. A rejection becomes a redirect.

What we learned

The best place to solve a sorting problem is before the sort happens. Every intervention that moved upstream — before the donor packs, before they drive over — had more leverage than anything after the item arrived. And Gemini 2.5 Flash is genuinely capable at condition assessment from a phone photo with the right prompt structure.

What's next for DonateSmart

Closing the donor loop — notifying them when their item arrives and sells, with real mission impact numbers. Demand broadcasting — staff publish what's needed this week, donors see it before they pack. Surge forecasting — combining calendar intelligence with live pre-registered data to give staff a 7-day volume prediction. And item passports — making every QR code a public story buyers can scan at the shelf.