Kickback

The AI bookkeeper that finds the money you're leaving on the table.

Comment

Inspiration

Every year, tens of millions of self-employed people overpay their taxes - and they don't even know it. Freelancers, creators, gig workers, solo founders. Not because they want to, but because no one is watching their expenses. A human bookkeeper costs $300–600 a month, so most independent earners simply go without. The result: according to IRS Publications 463 & 535, the average self-employed person misses $3,000–$8,000 in legitimate write-offs every single year, and roughly 70% miss deductions entirely. That's money that is legally theirs, quietly lost - because the tax system is built for people who can afford professionals.

I built KICKBACK to close that gap. To give every independent earner the bookkeeper they could never afford - for the price of a coffee.

What it does

You upload your business transactions - a CSV bank export, or a quick paste. KICKBACK's AI reads each one the way a sharp accountant would: it identifies the merchant behind a cryptic string like AMZN MKTP US*2X4T9, sorts it into the correct tax category, and makes the judgment call on whether it's a legitimate, deductible business expense. Within seconds, you see exactly how much you're owed back - itemized, explained, and ready for tax time.

No accounting knowledge. No bank login. No $300 bill. Just $2 a month.

How we built it

KICKBACK is a full production fintech application:

AI engine: Google's Gemini API reads and reasons over every transaction, returning structured JSON - merchant, category, deductibility, confidence level, and plain-language reasoning for each line. Frontend: Next.js 16 (App Router, TypeScript), deployed on Vercel. Auth & database: Supabase - email/password + Google OAuth, Postgres with row-level security, persisting every analysis and the user's lifetime stats. Payments: a complete freemium subscription flow - free scan, $2/month Pro, paywall, and billing.

The whole loop works in production: land → sign up → onboard → upload → AI analyzes live → see your found money → save history → subscribe.

Challenges we ran into

Building a complete, credible fintech product solo - auth, database, a real AI engine, and payments - is a lot of moving parts, and every layer had to actually work, not just demo. Payment activation was the sharpest challenge: as a solo founder in Nigeria, live card-payment activation requires business-registration documents I don't yet have. Rather than fake revenue, I built the full monetization flow and kept it honest - the model is complete and integrated, and KICKBACK is launching to its first users now.

Accomplishments that we're proud of

  • Built a complete, production fintech product solo, from zero - real auth, database, AI engine, and payments, all live.
  • The AI makes real judgments, not lazy ones - it correctly rejects a solo coffee as personal while claiming a client flight, the exact call a bookkeeper is paid to make.
  • Made the AI the whole operation, not a feature - Gemini reads and judges every transaction end to end, so one person can serve thousands.
  • It looks like a real company's product - a premium, cohesive fintech interface across every screen.
  • It's live and works right now at kickback-dun.vercel.app, with Gemini running real analyses in production.
  • Turned a documented problem into a real fix - giving self-employed people the bookkeeper they could never afford, for $2.

What we learned

That the AI is not a feature you add - it can be the entire operation. The hardest and most rewarding part was engineering the deduction-reasoning prompt so the AI makes real judgment calls, not lazy ones. It correctly rejects a solo Starbucks as personal while claiming a client dinner. That single distinction - refusing to inflate the number - is what makes it trustworthy. I learned that building an AI-native business means designing the human role around what the AI can't do: quality, trust, and edge cases.

What's next for Kickback

Real customers. The product is live, the AI runs in production, the model is built. The next chapter is putting KICKBACK in front of the independent earners who need it - and expanding the deduction logic to every country's tax code.

Built With

Share this project:

Updates