The problem Blocks And Scissors solves

Blocks & Scissors solves the trust problem that happens when a customer and a barber disagree over whether the final haircut actually matches the reference that was agreed on beforehand. Normally, the customer has to either pay despite being unhappy, or refuse payment and leave the barber with no guaranteed protection.

Our solution uses an AI-verified escrow system. The customer locks the payment in a Sui smart contract before the service. After the haircut, the reference image and final result are compared by our AI scoring pipeline. If the score meets the agreed threshold, the barber is paid automatically. If it does not, the customer is refunded.

This makes the process safer for both sides: customers get protection against clearly mismatched results, while barbers get protection against customers refusing to pay after an acceptable service. The same model could later be applied to other reference-based services such as tailoring, tattoos, detailing, repairs, and commissioned work.

Challenges we ran into

One of the biggest challenges was getting the AI score to react to the haircut itself rather than differences in the photos. Raw image similarity was too sensitive to lighting, backgrounds, camera angle and styling. I improved this by adding hair-focused crops, grayscale and blurred shape comparisons, mirrored side-profile matching, and separate checks for haircut family, texture, back length and fade/taper structure. I also added confidence-aware penalties so uncertain classifications do not affect the score too aggressively.
Another challenge was connecting the off-chain AI result to an actual on-chain payment. The AI runs in the FastAPI backend, while the escrow lives on Sui. I built an oracle flow where the backend submits the final score through an authorized OracleCap, and the Move contract decides whether to pay the barber or refund the customer. I tested the flow using Sui dry-runs and live testnet transactions, and also handled issues such as network mismatches, gas configuration and preventing an escrow from being settled twice.
The final MVP works end-to-end, but the AI is still optimized mainly for side-profile fade and taper comparisons, so broader hairstyle support and multi-angle verification would be the next major improvement.