Walmart Search & Retail Knowledge Graph
Search at Walmart is a hundred-to-a-hundred problem — a mature system where the work is squeezing reliable basis points out of enormous volume. I worked on relevance and the retail knowledge graph underneath it.
At Walmart’s scale, small relevance gains move enormous absolute volume — but the search stack lacked the structured understanding to make those gains reliably.
When you operate at this scale, basis points are the product.
Built variant grouping into the result surface so shoppers chose between products, not between near-identical SKUs — less cognitive load, higher conversion.
Introduced a session intent graph so the system reasoned about what a shopper was trying to do across queries, not just the current string.
Invested in non-English understanding, which turned out to be a large untapped pool of real purchase intent.
Mature-system product work is emotionally different from 0→1 — there’s no launch high, just basis points that compound into real money. The skill is believing the small numbers matter and having the measurement discipline to prove they do. Two USPTO patents came out of this work on search and item identification.