How Jupyter Agent Blew My Mind. The AI Revolution You Didn’t See Coming.

Easily accessible, but hard to believe. Once you get your hands on it you will see what I am talking about!

I assume we have all heard about AI Agents. It is breaking into the mainstream and has been an emerging buzzword. I assure you though, after I myself have experimented with Jupyter Agent, it’s capabilities are not be brushed over.

What is Jupyter Agent?

The Jupyter Agent is an AI Agent in an HuggingFace space. It is able to creat a whole python notebook with a single prompt, with autonomous error handling, outputs and ability to read uploaded files. While it is called Jupyter Agent, it works the same way as Google Colab which you some of you be more familiar with for creating python notebooks.

It is online, free and accessible to all with no log-in! Check it out here!

“Agents are not only going to change how everyone interacts with computers. They’re also going to upend the software industry, bringing about the biggest revolution in computing since we went from typing commands to tapping on icons.” — Bill Gates, Co-founder of Microsoft

Let’s test it out!

One cookie-cutter project, that I also had to go through in University, was analyzing the Iris Flower Dataset. It is basically a dataset of Iris flowers, which are separated in three classes. Along with the target class, you are given the measurements of the flower to perform EDA and clustering.

Let’s see how much time this would have saved me back in the day! :D

A scatter plot EDA done in a matter of seconds.

As a student this took me hours… And yes I was even using ChatGPT.

An agent is able to create a bigger notebook projects, including data cleaning, data visualization and add predictive machine learning in a single prompt!

The link to the agent: https://huggingface.co/spaces/data-agents/jupyter-agent

You might be asking at this point:

Nothings new. How is this any different from just using ChatGPT on my projects?

1. Output displayed.

A really convenient feature missed in ChatGPT is seeing the output.

It so much easier to see the outputs and plots, as you usually have to copy, go to your IDE, paste the code, wait for it to run and its just spews garbage and errors.

2. Error Handling.

I mentioned that it displays outputs and errors if there are any. Sweet, but that’s not all.

If the agent does encounter an error, it will go back and adjust the code to fix the errors in one go!

3. Upload Files.

In the most effortless way, you can upload the file.

An example as in most cases, you upload a dataset and the agent reads it to do any task you ask it to with that dataset. EDA, Predictions, Cleaning. You name it — or prompt it, shall I say — and it does it!

4. Download the .ipynb file.

No more copying and pasting!

You get a one-click button to download the file.

5. System prompt configuration.

If you need to adjust the way the LLM responds, you can do that to.

The agent comes with a pre-written system prompt to instruct how it should respond and you can look through to adjust it to your liking!

What do you think?

I personally feel like I have underestimated agents and once I found this it made me see much clearer of all the capabilities. I realized many might be in the same bubble as I was, or maybe not.

Of course I might not move to Jupyter Agent to do all of my notebooks. I see myself using it to create a first draft boilerplate to add code myself. Then I would definitely use ChatGPT to get ideas and adjust the code.

This was more so an epiphanic experience to see the true power of agents and the ways it would be much more useful than an LLM, by practically using it first-hand.

I hope you have fun with this new agent and you found this article some what interesting.

Thanks for reading :)

Follow me after clapping and commenting here: https://medium.com/@savvas.theo1/subscribe