Having an AI project portfolio is no longer a differentiator: it's the new minimum requirement to access tech roles in Argentina. In a market where employers receive dozens of similar CVs, the profiles that show real projects with functional code and measurable results have a concrete advantage.

This guide explains exactly which projects to include, where to publish them, and how to present them so that a recruiter or technical lead understands your value in less than two minutes.

Why the AI portfolio is different from the traditional dev portfolio

A classic web development portfolio shows that you can build interfaces and connect APIs. An AI portfolio needs to demonstrate something more complex: that you understand the business problem, that you chose the right model to solve it, and that you achieved a measurable result.

This means it's not enough to upload a Jupyter Notebook with a trained model. The recruiter needs to see the context (what problem it solved), the technical solution (what tools you used and why), and the impact (what improved, by how much, with what data).

According to McKinsey's State of AI report, the demand for profiles with AI implementation skills grew more than 35% in 2024 in Latin America. Demonstrable projects are the main filter in selection processes.

Which projects to include in your AI portfolio

1. A RAG (Retrieval-Augmented Generation) application

Build a chatbot that answers questions about a set of documents (PDFs, web pages, knowledge bases). Recommended technologies: LangChain or LlamaIndex for orchestration, ChromaDB or Pinecone as a vector store, and any LLM via API (OpenAI, Anthropic, or Hugging Face models). This type of project demonstrates that you understand how context works in LLMs and how to avoid hallucinations.

2. An autonomous agent with tools

Develop an agent that can use external tools: search the web, read files, execute code, or query APIs. LangGraph, CrewAI, or the Anthropic SDK are good starting points. The differentiator lies in showing that the agent makes coherent decisions and handles errors without human intervention.

3. An ML pipeline with real data

Train a classification or regression model on a real dataset (Kaggle, open government data from Argentina, or your own data). What matters is not the final accuracy: it's the documentation of the process. Why did you choose that algorithm? How did you handle class imbalance? What did you do with the null values?

4. A fine-tuning or advanced prompting project

If you don't have experience training models from scratch, show expertise in structured prompting or fine-tuning a small model (Llama, Mistral, Qwen). A project that demonstrates you can adapt an LLM to a specific domain with concrete examples is worth much more than a well-configured generic model.

5. A dashboard or internal tool with AI

Build something that solves a real problem: an automatic report generator, an email classifier, a customer service assistant. Streamlit or Gradio let you create functional interfaces quickly. This type of project demonstrates that you can take AI to production.

Where to publish your projects

The first destination is GitHub. Each project should have its own public repository, with a clear README and frequent commits that show the development process, not just the final result.

For projects with interactive demos, Hugging Face Spaces is the standard platform of the AI ecosystem. You can deploy Gradio or Streamlit applications for free and share the link directly on your CV or LinkedIn. You can also publish trained models on the Hugging Face hub, which adds visibility within the technical community.

If you built something with Streamlit, you can use Streamlit Community Cloud to have a public URL without needing to configure infrastructure.

When building the demos, tools like Cursor AI can significantly speed up development, especially if you're learning to work with LLM APIs for the first time.

How to write the README to impress recruiters

A good AI project README should have this structure: one line that explains what the project does, the business or technical problem it solves, the tech stack with a sentence about why you chose each tool, the installation and usage instructions in fewer than five steps, and a GIF or screenshot of the final result.

Avoid generic READMEs copied from tutorials. Recruiters and technical leaders detect them immediately. A specific and honest README —even if the project is simple— generates more trust than a grandiose one about a mediocre project.

The most common mistakes when building an AI portfolio

• Only replicating tutorials: the "Hello World" MNIST classifier no longer impresses anyone. You need your own twist: different data, a specific domain, or an original question.

• Not documenting the process: the code alone doesn't tell the story. The commits, the comments, and the README are equally important.

• Overly ambitious projects that are never finished: a simple, complete, and working project is worth more than three half-finished projects.

• Ignoring the user experience: if the demo requires ten steps to work, no one will try it. The first impression matters.

Recommended Coderhouse courses

If you want to build the projects you just read about and need the technical knowledge to do it, Coderhouse has specific courses for every level:

• Introduction to Artificial Intelligence: fundamentals of ML and LLMs, ideal as a base before tackling more complex projects.

• AI Agents: learn to build autonomous agents with tools, the most in-demand type of project in AI portfolios.

• AI Engineering: to make the leap to production projects, with RAG, fine-tuning, and deployment.

Frequently asked questions

How many projects do I need to have in my AI portfolio?

Three solid, well-documented projects are enough to start applying. Quality always beats quantity. It's better to have three finished projects, with a complete README and a functional demo, than ten half-finished projects or ones copied from tutorials without modifications.

Do I need to have worked at a company to have real projects?

No. Personal, academic, or practice projects count if they demonstrate that you solved a real problem. You can use public datasets from Kaggle, open government data, or even data from a well-defined hypothetical situation. What matters is that the problem is clearly stated and the solution is functional.

Should I include projects that don't work perfectly?

Yes, as long as you're transparent about the limitations. In the README you can have a "Limitations and next steps" section where you explain what you would improve with more time or resources. That demonstrates critical thinking, which is one of the most valued skills in AI profiles.

Does the programming language I use matter?

Python is the de facto standard in AI and machine learning. If your projects are in Python, good. If you use another language (TypeScript with the Vercel AI SDK, for example), make sure to explain why. What matters is that the code is legible, well organized, and has clear logic.

How do I include my portfolio in my CV?

Include the link to your GitHub profile as the first contact detail. Below each experience or project in the CV, add a direct link to the specific repository. On LinkedIn, use the "Featured projects" section to link the Hugging Face Spaces or Streamlit demos. A short 60-second video showing the project can also make a difference in remote selection processes.