The kitchen
How long will it take to cook this?
Paste a recipe, or pick a card, and a classifier trained on 40,000 Food.com recipes sorts it into quick (under 35 minutes), medium (35 minutes to 4 hours) or slow (more than 4 hours). It also shows which words swayed it.
Start from a recipe card
Recipe card
Your recipe
The browser model runs locally, and nothing you type leaves your browser unless you ask for the optional LLM second opinion below, which sends it to the provider you chose.
Loading the model.
Fine print
What is running here
The 2021 models read staff-made Doc2Vec vectors or a 200-tree Random Forest, which are too heavy to ship to a browser. So this page runs two light companions trained on the same recipes and labels: TF-IDF word and word-pair weights feeding either Logistic Regression or Multinomial Naive Bayes. Both are written out to a small JSON file and re-implemented in TypeScript, and a test suite checks they give the same probabilities as scikit-learn.
The tokeniser drops single characters, so “bake for 5 minutes” reaches the model as “bake for minutes”. That pair turns out to be one of the strongest signs of a quick recipe. Words such as “hours”, “overnight” and “slow cooker” pull the other way. Slow recipes are rare (about 5% of the data), which is why 3 of the 4 slow cards (“Bircher overnight oats”, “Summer berry pavlova” and “Country sourdough loaf”) fool both models, just as class 3 tripped up every model in the report.
Want the original lineup? Compare the 2021 models or see why stacking did not help.