Inspiration

Crowdfunding sites ask donors to trust a story they cannot check. Creators with real needs often struggle to write a convincing campaign, and donors have no quick way to judge whether a campaign is realistic. HelpFund uses AI to help on both sides and uses PayPal to keep the money flow verifiable.

What it does

  • AI story writer: a creator enters rough notes and Google Gemini turns them into a structured campaign story with an itemized milestone budget.
  • AI trust audit: Gemini checks budget itemization, beneficiary clarity and timeline realism and produces a public 0-100 credibility score that donors can read before giving.
  • Impact estimate: the AI estimates what a given donation amount can achieve for the cause.
  • AI campaign images: the app generates a campaign visual, with fallbacks so a campaign never ends up without an image.
  • PayPal donations: donors pay through PayPal. The server creates the order, captures it after approval, and only then updates the campaign balance in the database.
  • Explore and dashboard: donors can browse and filter campaigns by category, sort by funding or trust score, and creators can track their campaigns.

How we built it

  • PayPal: Orders v2 REST API in the PayPal sandbox. The Node.js server gets and caches an OAuth 2.0 access token, creates orders, and captures payments server-side. The browser never decides how much was paid.
  • Ledger: donations are saved in SQLite (better-sqlite3, WAL mode) inside a transaction, with an idempotency check so the same PayPal order cannot be counted twice.
  • AI: Google Gemini for the story, milestone, impact and trust-audit features. Image generation uses Cloudflare Workers AI (FLUX.1 Schnell) and Google Imagen 3, with a fallback illustration if both fail.
  • Frontend: plain HTML, CSS and JavaScript served by the same Node server.
  • Hosting: deployed on Render.

Challenges we ran into

  • Making donations safe: the amount and status have to come from PayPal's capture response, not from the client, and a repeated callback must not double-count.
  • Getting dependable JSON from the AI so the trust score and milestone budget always render.
  • Keeping image generation working when one provider is slow or fails.
  • Cold starts on a free hosting tier.

Accomplishments that we're proud of

What we learned

How the PayPal Orders v2 create/approve/capture flow works end to end, how to design a small idempotent ledger, and how to turn free-form AI output into structured data the UI can rely on.

What's next for HelpFund

Real campaign verification beyond the AI score, creator payouts, donor receipts and notifications, and a production PayPal account after proper review.

Share this project:

Updates

Submission history