Inspiration
What it does
How we built it
Challenges we ran into
Accomplishments that we're proud of
What we learned
What's next for Fieldproof
Inspiration
Every neighborhood has a project that never happens. A school workshop that needs benches. A garden that needs a fence. People are willing to chip in, but nobody wants to be the one holding the cash, and nobody trusts a spreadsheet they cannot see.
The money is not the hard part. Trust is. I wanted to build something where a volunteer with no finance background can raise money, spend it, and prove it, without anyone having to take their word for it.
What it does
Fieldproof takes a community project from idea to receipt.
- An organizer uploads a rough plan, a sketch and a few quotes. AI turns that into a clean blueprint with fundable packages, like "eight electronics benches".
- Neighbors pledge through PayPal. Nothing is charged until the goal is met. When it is, each pledge is collected on its own.
- If one pledge fails, the round is not lost. The donor gets a PayPal invoice with a QR code and 48 hours to complete it. In the demo, a $35 pledge is recovered this way and the round closes at $960.
- Volunteers buy supplies and photograph the receipt. AI reads it, checks it against the plan, and flags duplicates.
- The organizer approves, and the volunteer is repaid through PayPal Payouts to a verified PayPal account.
- Everything ends up on a public Proof Page with a QR poster for the wall, and a printable Board Pack for the committee.
- Each donor gets a personal trail: here is your $50, and here is the bench it helped buy.
How we built it
- Next.js on Vercel, Postgres for the ledger, private file storage for documents and receipts.
- PayPal Orders for checkout, Vault for conditional pledges, Invoicing for recovery, Refunds for rounds that fall short, Payouts for reimbursements, Log In with PayPal to verify who gets paid, and Webhooks to keep state in step.
- The PayPal layer is built on the SDK generated by APIMatic.
- AG Grid's AG Studio powers the organizer workspace, with two AI agents on top: a Steward that explains the money and an Auditor that looks for problems.
- OpenAI models handle planning and writing. Gemini reads receipts and layouts.
- English and Spanish, with ready made templates for common project types.
How AI uses PayPal
AI reads the plan, proposes the packages, extracts receipts and drafts updates. It can look up real PayPal orders and invoices through the PayPal Agent Toolkit to answer "where did this money go". It never charges, refunds or pays anyone by itself. Every money action is approved by a person and confirmed against PayPal before it shows as done.
Tools used and how
- PayPal: Orders, Vault, Invoicing, Refunds, Payouts, Log In with PayPal, Webhooks, Agent Toolkit (sandbox)
- APIMatic: generated PayPal SDK used for the whole payment layer
- AG Grid (AG Studio): organizer workspace, widgets and reporting
- OpenAI and Google Gemini: planning, receipt reading, duplicate detection, writing
- Vercel and Postgres: hosting, scheduled jobs, ledger
Challenges we ran into
Conditional money is tricky. A pledge is a promise, not a payment, and the product has to say that clearly to donors. We spent a lot of time on wording and on what happens when things go wrong: a declined card, a round that falls short, a receipt submitted twice.
Receipts were the other challenge. Real receipts are crumpled, cropped and photographed at an angle. We ended up combining exact file checks, image similarity and text matching so that a duplicate is caught without accusing an honest volunteer.
Accomplishments that we're proud of
- A cold visitor can watch a whole project get funded, spent and proven in under a minute.
- A real sandbox run collected eight separate pledges totaling $960 and paid a volunteer back through Payouts.
- Every number on the public Proof Page traces back to a PayPal record or a reviewed receipt.
What we learned
People give more easily when they can see the end of the story. Building the proof first, and the fundraising second, changed the whole design.
What's next
Pledges for any package, matching grants from local sponsors, and a pilot with a real school or neighborhood group.
Notes for judges
The 60 second tour runs in a private Sample workspace created just for you, and Sample data is marked as such. The README lists which flows were also run against the live PayPal sandbox.
