The pantry of every review
Every dish.
Every version.
Millions of scattered reviews, folded into one map of how the world actually eats — version by version, plate by plate.
The problem
A million voices,
one kitchen
Hundreds of millions of food reviews exist online — scattered across Google, Yelp, TripAdvisor, Reddit threads, and independent food blogs.
They're indexed by restaurant. Never by dish version. A review mentioning "the biryani" at a restaurant serving three different regional styles tells you almost nothing on its own.
Polydish reads them anyway — matching each mention to the exact version of the dish it's actually describing.
How Polydish works
Raw data, understood
Raw reviews. Millions of reviews, scattered across Google, Yelp, TripAdvisor, Reddit, and food blogs.
Dish extraction. Which dish. Which version. Matched automatically from what the review actually says.
Sentiment synthesis. Recurring praise and complaints, scored and pulled into patterns — by a language model, not a person.
The write-up. A synthesis paragraph, a pull quote, and what actually makes this version distinct.
The scale of taste
What's actually in the pantry
265,080
Reviews synthesised
154
Dish versions mapped
35
Cuisines covered
Real numbers, counted from the current catalog — not rounded up for effect.
Our mission
Taste without borders
Food carries memory and identity as much as flavor. An Indian expat in Chicago craves the exact biryani their mother made — not "biryani" in the abstract. Polydish exists so the specific version you're homesick for is findable, wherever you are.
Every write-up here is AI-generated from real review text, not hand-fact-checked line by line. Full methodology · Spot something off? Tell us.