World Flavor Atlas. Radzikowski and Chen's food embeddings in 3D

Name: World Flavor Atlas. Radzikowski and Chen's food embeddings in 3D
Availability: InStock
Author: drinkcocacola

by drinkcocacola·May 27, 2026·2 points·0 comments

Visit Project View on HN

AI Analysis

●●SolidEye CandyBig Brain

Real-time slider between three embedding models shows how ingredients cluster differently.

Strengths

•Slider reshaping clusters across COOC, CORE, and CHEM models is genuinely clever interaction design.
•Makes abstract ML embeddings tangible for non-technical audiences exploring food science.
•Based on real research with 4.14M recipes across 7 languages, not synthetic demo data.

Weaknesses

•Limited to 250 ingredients feels more like a demo than a complete exploration tool.
•No clear use case beyond education and exploration — not actionable for chefs or researchers.

Post Description

After reading the paper (https://arxiv.org/abs/2605.22391), I started playing with a visualization that helped me understand it more deeply. This is the result. Hope you enjoy it!