2026-04-24

How Aji IQ Scores Every Dish: The Science Behind the Numbers

Aji IQ Team|5 min read

We score every dish by analyzing menus, Google ratings, and community signals — then match the result to your taste profile.


Most food apps score the restaurant. We score the plate. Every dish on Aji IQ gets a single number from 10 to 98 that answers one very specific question: is this particular thing worth ordering?

Menus first, not reviews

The score starts with the menu. When a restaurant gets added to Aji IQ, we pull its full menu from Google, delivery platforms, and direct publisher feeds and turn every item into a dish record — no review required. That means new restaurants and obscure specials show up from day one instead of waiting months for crowd coverage to catch up.

The composite score

From there we blend three engines. The taste engine compares each dish against a 128-dimension flavor vector we learn from your five-screen onboarding and every dish you have saved, skipped, or checked into. The quality engine folds in the venue's Google rating, a silent Bayesian signal trained on hundreds of thousands of independent reviews, and rolling check-in counts. The crowd engine reads micro-feedback — chips, claps, and short notes left by other diners — so popular dishes rise even when the crowd disagrees with a restaurant's overall rating.

Confidence, not false precision

Raw math is not the same as confidence. Every score ships with one of three tiers: Established dishes (solid border) have enough signal that the number is stable, Growing dishes (dashed border) are promising but still settling, and New dishes (dotted border) lean harder on priors and menu content. We render that tier right next to the ScoreArc so you never mistake a lucky sample size for a real pattern.

What the score is not

The score is not a star rating in disguise, not an ad slot, and not a vote. A dish with 2 glowing early reviews is not automatically better than a dish with 200 honest ones — our Bayesian shrinkage pulls low-data dishes toward the sensible prior and lets high-data dishes speak for themselves. The number you see is the best honest answer we can give, updated daily.