natural / README.md
Wayfinder6's picture
add demo + social links
9f0159b verified
|
Raw
History Blame Contribute Delete
6.43 kB

A newer version of the Gradio SDK is available: 6.20.0

Upgrade
metadata
title: Natural
emoji: 
colorFrom: green
colorTo: yellow
sdk: gradio
app_file: app.py
license: mit
short_description: Pick a food. See how far it's come.
tags:
  - build-small-hackathon
  - track:backyard
  - achievement:offbrand
  - achievement:fieldnotes
  - achievement:sharing
  - gradio

Natural -- Pick a Food. See How Far It's Come

Wild bananas had seeds the size of peppercorns and would break your teeth. Wild almonds would kill you -- actual cyanide. The corn on your plate started as a grass with a cob the size of your pinky finger. Every food you eat is a 10,000-year engineering project, and the label says "natural."

Name a food. This app shows you what it used to be.

What it does

You type a food -- any food, from corn to chicken to strawberries -- and the app traces it back to its wild ancestor. What it used to look like, what size it was, what it tasted like, and the specific human interventions that turned it into the thing on your plate. It names eras, regions, and techniques: selective breeding, hybridization, grafting, industrial processing. It doesn't take a side on whether the changes are good or bad. Some saved a billion people from famine. Some are about shelf life. Both are part of the story.

How it works

The input goes to Mistral-7B-Instruct-v0.3 via the HuggingFace Inference API with a system prompt that enforces a 5-7 sentence format. The prompt tells the model to show what the original wild version looked like with specific details (size, color, taste, texture), name the key moments of human intervention, and include one thing about the modern version that would be unrecognizable to someone who only knew the original. The tone is "a friend who just went down a Wikipedia rabbit hole about corn at 2am." Temperature is 0.8. If the API is unavailable, the app falls back to its handwritten response library. Unrecognized foods get a generic response that still frames the user's input within the broader story of human food engineering.

The fallback system

The app ships with 9 handwritten fallback responses covering corn, bananas, watermelon, carrots, chicken, strawberries, almonds, tomatoes, and wheat. Each one was researched and written to include specific historical details that most people have never encountered. The banana fallback explains why artificial banana flavor doesn't taste like banana -- it's based on a variety that went extinct. The almond fallback reveals that wild almonds are genuinely poisonous and every almond you've ever eaten descends from mutant trees. The tomato fallback explains that the gene making grocery store tomatoes uniformly red also suppresses the gene that makes them taste good. These responses work as standalone micro-essays. They don't need a model behind them. They are the product.

Why this matters

The word "natural" is the most abused word on every grocery store label. It means almost nothing, legally or scientifically. But people use it as a value judgment -- natural is good, unnatural is bad -- without knowing that the line between the two disappeared about 10,000 years ago when humans started selecting which seeds to plant. This app doesn't preach about GMOs or organic farming. It just shows you the distance between the wild original and the thing in your kitchen. The tone is wonder, not warning. Humans have been engineering food since before they had a word for engineering. That's not a scandal. It's one of the most impressive things our species has ever done.

Try it

  • "corn" -- started as a grass called teosinte with a cob the size of your pinky, 9,000 years of breeding made it unrecognizable
  • "bananas" -- every Cavendish banana is a sterile genetic clone, and we already lost one variety to a fungus
  • "almonds" -- wild almonds contain cyanide, every almond you've eaten descends from a mutant tree
  • "chicken" -- a 1948 contest called "Chicken of Tomorrow" redesigned the bird from a 2-pound junglefowl to a 6-pound broiler in 42 days

Built with

  • Gradio
  • Mistral-7B via HuggingFace Inference API
  • Python
  • Zero external dependencies beyond gradio and huggingface_hub

Process notes

This app required more research per fallback than any other app in the suite. Every handwritten response had to include verifiable historical claims -- specific dates, specific regions, specific biological mechanisms. The corn entry references teosinte and 9,000 years of Mesoamerican breeding. The wheat entry references Norman Borlaug and the Green Revolution. The chicken entry references the 1948 "Chicken of Tomorrow" contest. These aren't generated facts -- they were individually researched and cross-checked. The system prompt had to be engineered to produce the same level of specificity from the model, which meant including explicit instructions to "name the era or region when possible" and "show what the original wild version looked like -- be specific."

The design philosophy behind the fallbacks is the same as the rest of the suite but the execution is different. The other apps deal in feelings and interpretations. This one deals in facts. A wrong fact about corn's origin isn't just a bad response -- it undermines the entire premise. So the handwritten fallbacks aren't just well-written, they're accurate. The banana entry about Gros Michel and Panama disease is real. The carrot entry about the Netherlands and beta-carotene is real. The tomato entry about the uniform ripening gene suppressing flavor is real. Accuracy is part of the voice. You can't say "here's what you didn't know" and then get it wrong.

The UI uses a green gradient to signal the subject matter immediately -- food, plants, nature. The button says "Trace it back" because the app is a time machine for your grocery list. Clickable example buttons let users start with the most surprising entries (corn, bananas, almonds) before exploring on their own. The light theme and clean typography serve the same purpose as in every app in the suite: no friction between the user and the content. The app loads fast, works immediately, and the first response is worth reading. Everything else follows from that.

Demo & Social


Heuremen -- Let's stay connected and keep information free.