<- all tokdocs

Vibe Coding on Hard Fork: Flavor Finder and Ro-Sham-Toe

Watch on TikTok

View on TikTok ->

A caller named Matt Holman, a graphic designer from Los Angeles, showed off two vibe coding projects on the Hard Fork podcast. One is a genuinely useful cooking tool. The other is an intentionally bad game that doubles as an LLM benchmarking exercise.

Flavor Finder: A Color Palette Generator for Ingredients

Matt built Flavor Finder with Claude. The app combines the dataset from The Flavor Bible (a well-known cooking reference that maps ingredient pairings) with the UI pattern of a color palette generator.

Hard Fork podcast episode card showing "Vibe Coding: Flavor Finder & Ro-Sham-Toe" with hosts at desk

You can browse ingredients one by one and build a palette, or hit generate to create a five-ingredient combination where everything pairs well together. You can lock specific ingredients and substitute others, shuffle the arrangement, and explore combinations you would never think of on your own.

Flavor Finder app interface showing ingredient search with tags like basil, garlic, leek, and a generated palette of fennel seed, crawfish, rice, tomato, and parsley

The app includes taste profile data for each ingredient, with a breakdown by umami, sweet, salty, and fat scores.

Flavor Finder detail view showing crawfish taste profile with umami, sweet, salty, and fat scores in a donut chart

Ro-Sham-Toe: The Worst Games Combined

Matt's second project is called Ro-Sham-Toe, a mashup of Tic-Tac-Toe and Rock Paper Scissors. He describes them as "the two worst games of all time" and asks various LLMs to combine them into something fun. His admission: it has struggled to produce anything genuinely entertaining, but the exercise is a useful way to observe how different LLMs approach creative problem-solving.

The Hard Fork hosts joked that the New York Times had acquired Ro-Sham-Toe for a billion dollars.

Key Takeaways

  • Flavor Finder shows how vibe coding can turn a reference book into an interactive tool by combining a known dataset with a familiar UI pattern
  • Using intentionally simple prompts (like combining two bad games) is a practical way to compare how different LLMs think
  • Claude was the LLM used to build Flavor Finder, handling both the data modeling and the UI generation

Resources

Published May 25, 2026. Writeup generated from a favorited TikTok.