Ash Mofidi

I wrote the first doc over a weekend. On Monday I found the three people who'd quietly wanted to build it for two years.

The problem was absurd: Walmart's international pricing was happening in spreadsheets. Category leaders would push an update and wait 24 hours for it to land in Canada.

We rebuilt the pipeline as a streaming system with ML-driven price adaptation. I owned the story across legal, operations, data science, and seven country teams.

Two sister teams copied the pattern. By the time I left for the next role, the ML roadmap was a team of its own.

The old pipeline was rules held together with regex and hope. The new one got a patent.

Brand classification at marketplace scale is really two problems. First, what is a brand? Walmart's catalog had 2.5 million inconsistent entries. Second, when someone lists a counterfeit, how do you catch it in under a second?

We paired embedding retrieval with an LLM judge that saw top-k candidates plus structured rationale. The model never saw raw identifiers, just normalized features. Every decision produced an audit trail a lawyer could read.

Filed with two engineers at USPTO. Patent pending. The real shipping metric: counterfeits down 40 percent in six months.

Twenty engineers, designers, and analysts across four time zones. One Gantt chart. A six-month deadline nobody thought we'd hit.

The technical work was never the hard part. The hard part was keeping legal and engineering using the same vocabulary for "risk." KYC, sanctions, FX, settlement timing, dispute flow. Each of those is a separate tribe with its own religion.

We hit five million GMV in ninety days. Seller onboarding went from six weeks to two days. The platform is how Walmart grows into Mexico and Canada now.