Building Data Science Projects Using AI: A Vibe Coding Guide

Image by Author

Vibe coding is one of the biggest AI trends in 2025.

If you haven’t already heard about it, vibe coding is essentially AI-assisted coding. You just describe what you want to build, and AI creates the entire application for you. If you run into errors, the model will fix them. It generates, tests, and debugs code with limited human intervention.

While many programmers have raised concerns about this trend, calling it a “risky shortcut” and pointing out valid risks like increased long-term technical debt, the “vibe coding” trend isn’t going away anytime soon.

And while I’ve seen a ton of hype around vibe coding in the software development field, I haven’t heard many data scientists talk about it.

As a data scientist myself, I “vibe code” almost every day at work. If used wisely, I think this technique can help you become a more efficient data scientist. In fact, if you are an aspiring data scientist, learning the right AI tools will help you stay ahead of the competition. You can bring your ideas to life by building creative portfolio projects — helping you stand out to potential employers. Along the way, you will also learn about new frameworks that will help you become a better data scientist.

In this article, I will explain how you can build data science projects with vibe coding.

Examples of Vibe Coding Data Science Projects

 
When I first started out in the field of data science, I built many portfolio projects and deployed them. Here is an example of a facial recognition app I created years ago:

Image by Author

I built a neural network that could predict a person’s celebrity lookalike from a photo they uploaded. This was one of the many portfolio projects that helped me land my first data science job.

The most time-consuming part of this project? Deployment.

It took me days to create the front-end web application — designing the app, allowing users to upload a photo of themselves, building the progress bar, and generating the output.

You see, like many other data scientists, I don’t know front-end programming languages like HTML and CSS. My focus is primarily on building models and analyzing data with Python. However, employers are no longer impressed by the Python code in your GitHub repository.

A front-end application like the one I showcased above is far more appealing, as it allows hiring managers and employers to interact with the models you have built. For aspiring data scientists, this is where vibe coding comes in. In just minutes, you can build a machine-learning model and get AI to develop a complete web application like the one above.

Here are two applications I created with AI in less than 5 minutes:

1. Twitter Sentiment Analysis Model

Image by Author

2. Titanic Survival Prediction Model

Image by Author

Let’s explore how you can build projects like the above to impress potential employers and get hired as a data scientist.

Building Data Science Projects with Vibe Coding

Step 1: Choosing an AI Tool

You can easily vibe code with tools like ChatGPT and Claude by explaining your requirements to the AI and pasting its output into your programming IDE.

However, I suggest going a step further and using an AI assistant that integrates directly with your IDE, such as Cursor AI, Lovable, and Copilot in Visual Studio Code.

These tools will access your entire codebase (datasets or any other files in your directory). They will then analyze your codebase, generate, and directly run code that meets your requirements.

I use Cursor AI at work almost every day, and it saves me a ton of time.

To get started with Cursor AI, you can go to this website, follow the installation instructions, and have it set up in seconds. You will then see a screen that looks like this with a chat box on the right, where you can type prompts.

Image by Author

If you’d like a more comprehensive guide on getting started with Cursor AI, I recommend watching this video.

Step 2: Generating Project Ideas

The next step is to ideate. You want to create a project that is unique — something that potential employers will find interesting.

If you find yourself stuck, you can type the following prompt into an AI chat interface to get some ideas:

Generate 5 unique data science portfolio project ideas that include both analytical components and interactive front-end dashboards. These projects should use non-traditional datasets.

Step 3: Preparing your Codebase

If you’re using an AI assistant like Cursor that has access to your codebase, then create a directory with all the project files and datasets required. For example, if you want to build a sentiment analysis model, your directory will comprise a single training dataset.

If your project is more complex, I suggest creating a separate file to document the following:

• A markdown file containing your project’s requirements. If you have a specific request, like ensuring the model avoids using a library that might be deprecated, you can add it here.
• For domain-specific code or lesser-known libraries, include documentation so the model doesn’t run into too many errors.
• Examples of code notebooks with similar workflows (this saves time because it gives the model context and tells it how to approach the problem).

Step 4: Creating the Prompt

Once you’ve prepared your directory with relevant files, you can start prompting the AI to give you code to build the project.

In Cursor, there is a chatbox on the right-hand side of the screen that allows you to do this.

Here are some prompting best practices when vibe coding:

• Provide a ton of context: To avoid running into errors, you must provide the model with as much context as possible. With tools like Cursor, you can even include an image of what you’d like the end product to look like.
• Force it to read your documents: When working with large codebases, I’ve noticed that AI assistants tend to skip over the files I uploaded and instead start to hallucinate column names. To prevent this from happening, you need to explicitly prompt the AI to read through specific documents before generating code.
• Role-playing: You can also ask AI to take on the role of a domain expert before building a project. This approach can help you generate richer output as it tells the model to draw from a subset of its knowledge base that relates to a specific domain.

Here is an example of a prompt I used to build a sentiment analysis web application with Cursor:

Create a sentiment analysis web app that:

1. Uses a pre-trained DistilBERT model to analyze the sentiment of text (positive, negative, or neutral)
2. Has a simple web interface where users can enter text and see results
3. Shows the sentiment results with appropriate colors (green for positive, red for negative)
4. Runs immediately without needing any training

Please connect all the files properly so when I enter text and click analyze, it shows me the sentiment results right away.

Step 5: Iterate and Refine

AI models like Cursor can handle smaller projects just fine but tend to hallucinate and run into errors when working with larger codebases.

This is where your domain expertise comes in; for instance, explaining to the AI how a certain metric should be defined, which dataset to pull from, or even telling it which libraries to use and what to avoid.

One approach I’ve found especially useful is switching between a tool like Cursor and a different LLM like Gemini 2.5 Pro. If you find Cursor running into the same error multiple times, it might be because you aren’t explaining your requirements comprehensively. You can copy and paste the error into a different LLM and get it to generate a comprehensive prompt, which you can then paste into Cursor.

This multi-LLM approach works well when I run into errors while vibe coding.

Vibe Coding for Data Science: Yay or Nay?

 
I personally don’t believe that you can vibe code your way to production-ready code.

If you are a data scientist, I still suggest learning Python and SQL; otherwise, you will end up with computationally expensive code that leads to long-term technical debt.

However, I think vibe coding has its merits for specific use cases, such as building a data science portfolio project if you have no front-end coding knowledge.

You can also use it to speed up your workflows and quickly learn new libraries and techniques you aren’t familiar with — as long as you remember core programming concepts and don’t get over-reliant on AI.

As a next step, I recommend familiarizing yourself with the following AI tools that will make you a more efficient data scientist:

• Cursor
• Github Copilot
• Lovable

Natassha Selvaraj is a self-taught data scientist with a passion for writing. Natassha writes on everything data science-related, a true master of all data topics. You can connect with her on LinkedIn or check out her YouTube channel.