Build your first AI model.
Then put it online.

A Pakistani built the box.

Abubakar Abid. Son of two Pakistani doctors, grew up in a small town in Wisconsin. In his PhD at Stanford he trained a model that reads heart scans, and the cardiologist he built it for could not run it. It was just code in a notebook.

So he built a web page the doctor could drop a scan into. That page became Gradio. His three roommates quit big tech to build it with him, and in late 2021 Hugging Face bought the company. He runs the Gradio team there now.

43,542 stars on GitHub1,000,000+ developers a monthfree and open source

Stars from the GitHub API, 16 Sep 2026. Monthly developers from Hugging Face's own post, April 2025. His story is from his thread of 11 Jan 2024 and the Paul & Daisy Soros Fellowship bio.

Four steps. About ten minutes.

Follow these in order. If you can open a terminal and you have Python installed, you can finish this.

pip install gradio transformers torch

from transformers import pipeline

clf = pipeline("sentiment-analysis")
print(clf("honestly this is the best thing i have built"))

3

Wrap it in five lines

Save this as app.py and run python app.py. A web page opens with an input box, a button, and the output, all built for you.

app.pyCopy

import gradio as gr
from transformers import pipeline

clf = pipeline("sentiment-analysis")

def check(text):
    r = clf(text)[0]
    return {r["label"]: r["score"]}

demo = gr.Interface(fn=check, inputs="text", outputs="label")
demo.launch()

That is the whole app. gr.Interface takes your function, what goes in, and what comes out, and builds the page from those three things.

This is that exact code running. Nothing added.

demo.launch(share=True)

gradio deploy

Change what it does.

The shape never changes. You swap the task in pipeline() and adjust what goes in and out. Same five lines.

pipeline("sentiment-analysis")     # text  -> label
pipeline("image-classification")   # image -> label
pipeline("summarization")          # text  -> text
pipeline("automatic-speech-recognition")  # audio -> text
pipeline("translation_en_to_fr")   # text  -> text

gr.ChatInterface(fn).launch()

You have already used this.

These are not small projects. Every one of them is a Gradio page.

Star counts read from each repository, 15 Sep 2026.

Two honest things.

This is not training a model. You are using one someone else trained, which is what almost every real product does too. Training your own from nothing is a different job and you do not need it to ship something people can click.

The free hosting is a small machine. A Hugging Face Space on the free tier runs on a basic CPU and sleeps when nobody is using it. It wakes up when someone opens the link. Fine for a demo, a portfolio piece, or a final year project. Not for a business with real traffic.