FRIENDZ × GOAT

What started as an e-commerce audit became a national promo campaign platform with AI receipt verification — and an ongoing partnership that keeps growing sideways.

Overview

FRIENDZ is a Polish brand with a physical store, an online shop, and GOAT — their own beverage sold nationally through convenience chains. They didn't hire me for a project; the work kept finding the next project. The through-line of this case study is how a single engagement compounds: an audit surfaces the real problems, a campaign proves the platform, and a platform becomes the argument for what comes next.

My Role

Audit, architecture, full-stack development, AI pipeline, security, 3D content. Solo.

The Work

The audit that started it

The engagement opened with a storefront and infrastructure audit: what's broken, what's costing money, what's blocking growth. Instead of a document that gets filed away, the audit became the shared to-do list the rest of the partnership has been drawing from ever since.

The GOAT campaign — receipts in, rewards out

The centerpiece: a national promotion where customers buy GOAT, scan their receipt, and enter a juried contest for invitations to an event at the FRIENDZ house. Simple to say, hard to run honestly at scale:

AI receipt verification. Every submitted receipt runs through a pipeline: Claude vision reads it, byte-level forensics checks whether the file was manipulated, perceptual hashing catches duplicates, and AI-generated fakes get rejected outright. Digital receipts from Żabka arrive as signed documents — the platform verifies the cryptographic signature instead of trusting the pixels.

A real admin backend. A verification queue with appeals, fraud flags, user management, winners, notifications, dashboards, a "retry AI" button, and an emergency pause switch — because a live national campaign is not the place to discover you can't stop the machine.

Built to survive. A circuit breaker keeps the platform accepting submissions even if the AI layer stumbles. 73 integration tests run against real Postgres and Redis. A pre-launch security and performance audit caught a CSV formula injection and cut roughly 600 GB of projected transfer by removing dead 3D assets before peak traffic.

The 3D layer

Product work in Blender and React Three Fiber feeds the brand's campaigns: a faithful 3D model of the GOAT bottle for social content, and interactive 3D on the campaign landing. Same partnership, different muscle.

Where it's heading

The campaign platform is deliberately more than a one-off: accounts, points-for-purchases mechanics, and reward flows are the foundation for FRIENDZ WRLD — a fan loyalty world where buying the product feeds seasons, tiers, and rewards. The pitch for that platform stands on something better than slides: infrastructure that already ran a national campaign.

Results

A national campaign delivered end to end — landing, verification pipeline, contest platform, admin backend — hardened by tests and a security audit before launch. And the less visible result: a client relationship that moved from "fix our shop" to "plan our platform," which is the difference between a vendor and a partner.

Key Takeaways

Fraud handling is product design, not an add-on — the moment real rewards are involved, the verification pipeline is the product. And on partnerships: the best sales document I produced for this client was working software from the previous engagement.