The Vision AI Blog

Methodology, data-quality fixes, and how the analysis actually works — written straight from the engineering.

Methodology · 2026-07

Why We Rebuilt Player Similarity From Scratch

Raw-metric clustering let Federico Valverde and Frenkie de Jong land in the same bucket. That's not a fluke — it's what happens when you cluster event counts instead of playing style.

Data quality · 2026-07

The Position Label Bug Nobody Catches

William Saliba, Virgil van Dijk and Gabriel Magalhães were all classified as full-backs by our own model. The bug wasn't in the players — it was in assuming a centre-back sits dead-centre.

Product · 2026-07

A Report Engine, Not a Template

Early versions gave every player the same fifteen sections — a centre-back's report included a shot map built on four attempts all season. The fix was to make the report decide what it should contain.

Glossary · 2026-07

Every Metric In Your Report, Explained Simply

No jargon. What xG, xT, PPDA, Field Tilt and every other number in a Vision AI report actually mean — for someone who doesn't watch football or read data for a living.