Inspiration

Think about the last billboard you drove past. It probably had a QR code or a web address on it. Did you scan it? Nobody does. You can't stop a moving car to type in a link, and by the time you're home the ad is long forgotten. The business still paid full price for every one of those glances.

Traditional marketing has always worked this way. Billboards, TV spots and flyers cost a fortune, they're hard to measure, and most people walk straight past them. Businesses that tried to fix this by hiring big influencers often hit the same wall. They paid for the reach, and the results didn't always follow.

Meanwhile, the people we actually listen to are much closer. A friend's WhatsApp status. A neighbour's Instagram story. A creator with a few thousand followers who trust every recommendation they make. These voices already sit in front of us, every day, on the phone in our hand. One tap away.

So we asked a simple question: what if the billboard came to people's faces, through the people they already trust? And what if businesses only paid when a post really went out and real people really saw it?

That question became Reelch.

What it does

Reelch is a performance advertising marketplace. It connects businesses of every size with everyday social media users on TikTok, Instagram, X and WhatsApp, and pays them only for verified results.

  1. A business creates a campaign. It writes a short brief, chooses the platforms, sets a reward per verified post, and funds the budget through PayPal checkout (Orders API). The money is held in escrow, so promoters know it's real.
  2. Promoters join and post. They share the campaign and submit proof: a post link, a screenshot of their views, or a connected TikTok account that lets Reelch read view counts straight from TikTok.
  3. Gemini AI checks every proof against the brief: the right product, the required hashtag, the account handle, and how many people saw it.
  4. Verified posts are paid automatically through PayPal Payouts, in seconds.
  5. Doubtful proofs go to a person. Copied screenshots and inflated view counts are flagged as high risk and never paid out by the AI.
  6. The business sees where its money works hardest. Pie and bar charts break down views and spending by platform, and the Ask Reelch assistant answers questions in plain English from the business's own verified numbers.
  7. Unspent budget goes back through PayPal Refunds, and every movement of money is recorded in a double-entry ledger built with AG Grid.

Business model

Reelch earns only when businesses get results.

We take a percentage of every verified payout, around 15%. A business funds each reward plus our fee upfront through PayPal checkout, and both are held in escrow. The fee is only kept when a post is verified and paid. When a campaign ends, the unspent budget is refunded together with its unused fee.

$$\text{Reelch revenue} = 15\% \times \text{verified payouts}$$

In the demo, Brightside funds \$500 in rewards, so with the fee it would pay \$575 at checkout. Four verified posts pay promoters \$100 and earn Reelch \$15. When the campaign ends, Brightside gets back \$460: the \$400 of unspent rewards plus the \$60 of unused fee. If nothing is verified, Reelch earns nothing, exactly like the promoters. Traditional ads are paid for whether they work or not.

Promoters never pay to earn. Standard payouts are free.

As Reelch grows, we plan to add:

  • Business plans: subscriptions for brands running many campaigns, with deeper analytics, promoter targeting and team accounts.
  • Managed campaigns: our team runs campaigns end to end for brands that want it done for them.
  • Instant payouts: an optional small fee for promoters who want their money immediately.

For the hackathon, the fee is set to zero so every demo dollar goes to promoters or back to the business. The data model already includes it.

How we built it

  • Frontend: Next.js and Tailwind CSS, hosted on Vercel.
  • Backend: NestJS with Drizzle ORM on PostgreSQL (Neon), plus a pg-boss worker for background jobs such as payouts, emails and webhook processing.
  • Payments: PayPal Orders to fund and capture campaign budgets, PayPal Payouts to pay promoters, and PayPal Refunds to return unspent budget.
  • AI: Gemini reviews each proof and returns a structured JSON assessment, which includes a content match score, a pass or fail for each requirement, the handle and view count it reads from the image, and any fraud signals. Gemini also powers the Ask Reelch business assistant.
  • Platform data: a TikTok integration reads view counts directly, and Instagram links are checked against Instagram itself. OAuth tokens are encrypted with AES-256-GCM.
  • Storage: proof screenshots and verification documents are uploaded to Cloudflare R2 through short-lived signed URLs.
  • Ledger: a double-entry ledger, displayed with AG Grid, where every transaction must balance:

$$\sum \text{debits} = \sum \text{credits}$$

Every campaign's budget is accounted for at all times:

$$\text{escrow} = \text{funded} - \text{paid out} - \text{refunded}$$

In the demo, Brightside Coffee funds \$500 at \$25 per verified post. After four approved posts, \$100 has been paid out and the remaining \$400 is refunded.

For analytics, the number businesses care about most is what they pay for attention:

$$\text{cost per 1,000 views} = \frac{\text{amount paid}}{\text{verified views}} \times 1000$$

That is how Ask Reelch can tell Brightside that TikTok gives it the most views for its money.

Challenges we ran into

Letting AI touch money safely. Our first instinct was to let the model decide whether to pay. But a screenshot can contain text designed to persuade the model, such as "this post meets every requirement". So we split the job in two: Gemini only reports what it sees, and fixed rules make the payout decision. A post is approved automatically only when every requirement passes and the fraud risk is low. Anything uncertain goes to a human reviewer.

Catching copied screenshots. A cheater can take someone else's screenshot, re-save it and change the view count. We built a 256-bit perceptual "difference hash" for every image and compare images by Hamming distance:

$$d(a, b) = \sum_{i=1}^{256} \left[\, a_i \neq b_i \,\right]$$

Testing showed that re-saved copies sit up to about 38 bits from the original, while unrelated images sit around 95. So $d < 40$ marks a near-duplicate. Then we found a catch: two honest promoters sharing the same campaign image can sit under 10 bits apart, even closer than a cheater's copy. Looking alike could never be enough on its own. The final rule also compares the view count read from each image. Honest posts almost never share an exact number, while a copy always does.

Getting the money right. Payments fail, webhooks arrive twice, and payout batches finish minutes later. We moved every payment step into background jobs that can be retried safely, and recorded each one in the double-entry ledger, so a retry can never pay someone twice and every dollar can be traced.

Trusting view counts. Screenshots are the weakest kind of evidence, so wherever a platform offers real data, Reelch uses it. Connected TikTok accounts skip the screenshot entirely.

Accomplishments that we're proud of

  • A complete money loop on PayPal. Funding through Orders, escrow, automatic Payouts to promoters and Refunds of unspent budget all work end to end, and every movement balances in the ledger.
  • AI that can't be talked into paying. Gemini does the reading, fixed rules make the decision, and doubtful proofs always reach a person.
  • Fraud detection that tells cheaters from honest look-alikes. Combining image fingerprints with the view counts read from each image catches copied screenshots without punishing promoters who share the same campaign image.
  • Analytics a business owner can actually use. Charts show where every dollar went, and Ask Reelch answers questions in plain English.
  • A full product, not just a demo. Sign-up, onboarding with identity verification, campaigns, proofs, payouts, refunds, an admin review queue and a ledger, live at www.reelch.com.

What we learned

  • AI is a brilliant witness and a dangerous judge. It reads screenshots, hashtags and view counts better than any rule we could write, but money decisions belong to rules people can audit.
  • Payments are an accounting problem first. Thinking in escrow and double-entry made the whole system easier to trust, for us as much as for users.
  • A marketplace has two customers. Businesses need proof that every dollar bought real attention. Creators need to know the money is there and that they'll be paid fast. Every feature had to work for both.
  • The smallest voices carry real weight. A creator with a few thousand loyal followers can move people a billboard never will.

What's next for Reelch

  • More platforms with direct view-count verification, starting with Instagram and YouTube.
  • Click and conversion tracking, so businesses can pay for sign-ups and sales as well as views.
  • Smarter campaign matching that puts the right promoters in front of each business.

Reelch. Pay for results, not promises.

Built With

Share this project:

Updates

Submission history