Skip to content
Tamas Demeter
DTC E-commerce/Solo build, delivered at handover

Counterfeits of Your Products, Spotted Automatically Twice a Week

An eight-figure direct-to-consumer brand sells entirely through its own channels, with no authorized resellers, so anywhere its product images, name, or ad copy show up for sale is almost always a copycat or counterfeit. I built a self-feeding, two-workflow system that finds where the brand is being copied, filters the noise with an AI classifier, archives each hit as legal-ready evidence, and hands the team a short, ranked shortlist to act on.

Role
Solo build: architecture, AI classification design, evidence-archival engineering, handover
Tools
n8n ,Claude ,Reverse-image search ,Web archival ,Notion
Client review

Tamas built us a brand protection system that scans Google, marketplaces, and reverse-image search for counterfeit listings of our products, then runs every hit through an AI review layer and logs the real ones to Notion. It runs on its own twice a week and flagged 20 genuine violations in the first production run. He documented the whole thing, handed it over clean, and stayed responsive the entire time. Strong automation skills and clear communication. Would hire again.

Kfir S. / DTC Founder, UAE

Watch the walkthrough.

Coming soon

The problem

01

Search is easy, judgment is the job

A single sweep returns hundreds of results: news articles, the brand's own pages, unrelated products with similar words, generic category listings. Buried in that is a handful of genuine rip-offs. Hand a busy team that raw pile and they stop reading it by the second week. Finding hits is trivial. Deciding which ones are real is the deliverable.

02

The brand kept tripping its own alarms

The brand's own storefronts and social pages match its own brand terms perfectly. A naive monitoring system flags the company as a violator against itself, and that noise drowns the signal. Every result also needs a different next move, since a counterfeit on one marketplace, a copied listing on another, and a lookalike ad on a social platform each have their own takedown route.

03

The evidence has a short shelf life

The infringers who matter take their listings down the moment they sense pressure. If the proof is not captured at the moment of discovery, a later takedown or legal step has nothing to stand on.

The solution.

Architecture diagram, click to zoom

01

A watchlist that maintains itself

A scanner reads the brand's own store and live ads every week and keeps the list of assets to protect current on its own. Products change and campaigns rotate, and the watchlist updates without anyone maintaining a spreadsheet.

02

A monitor that hunts by text and image

Twice a week, a monitor hunts the open web and the marketplaces against the watchlist, by text and by image. It catches lifted product images, name matches, and copied ad copy wherever they surface.

03

An AI classifier that excludes the brand's own channels

Every candidate is read by an AI classifier that scores how likely it is to be a genuine rip-off versus background noise. The brand's own channels are excluded up front, so the system never flags the company against itself. What survives lands in one reviewable list, ranked by confidence, each row naming the exact takedown route to use for that platform.

04

Evidence archived at the moment of discovery

Every flagged page is archived to a public web archive the moment it is found, so there is a timestamped snapshot for a takedown or legal step even if the infringer deletes the listing the next day. Archival is not bolted on the end. It runs on every hit, by default, as part of the catch.

The impact

What the system delivers

  • Two self-feeding workflows: a watchlist scanner plus a web and marketplace monitor
  • On one production sweep it took in 101 candidate results, filtered 81 of them out as false positives, and surfaced 20 for human review, 18 likely and 2 borderline
  • More than 80 percent of candidates filtered as noise before a human looks
  • A ranked, reviewable digest with the per-platform takedown route named on every row
  • Every flagged page archived as timestamped evidence, by default, on every hit

The client signal

  • The third system this founder trusted me to build, after rebuilding his hiring and onboarding operation and a run of advisory sessions
  • Closed with a five-star review: "The work has been done on the highest level possible. We are very happy and satisfied with the results."
101 → 20
One sweep, raw candidates cut down to a review-ready shortlist
80%+
Of candidates filtered as noise before a human looks
2x / week
Automated sweeps of the web and marketplaces, by text and image
5 stars
Founder review on a third repeat engagement

Have a similar problem?

Tell me what is going on and I will tell you what I would do about it. No obligation.