Building an AI Project Portfolio: What Convinces an Employer?

In February 2024 the Burning Glass Institute, together with Harvard Business School, published a report on skills-based hiring. It concluded that many companies had announced they were dropping degree requirements, but that lasting change in hiring practice remained "elusive" for most of them. The lesson for a jobseeker is not that a certificate is worthless, since the market has not yet made up its mind, but that the best thing you can do is make your competence open to inspection. That is the job of a portfolio.

What do employers ask for?

The World Economic Forum's 2025 report says AI and big data are among the fastest-growing skills, that two-thirds of the employers surveyed plan to hire people with specific AI skills, and that analytical thinking is the most sought-after core skill. Demand exists, then, and the question is how an employer can tell that you have the skill and have not merely heard of it. A certificate answers in part; a project that runs and can be explained answers the rest.

Few and deep beats many and repeated

Do not judge a portfolio by the number of its projects. One to three, each carrying an engineering decision you can defend, are worth more than dozens of exercises copied from tutorials. Whoever reads it will ask: why did you choose this model? How did you deal with the shortage of data? What failed in the first attempt? Depth answers these questions; quantity does not.

Choose a problem you know from your own surroundings: a service whose setting you understand, public data in a field that interests you, or Arabic language processing. Stating the effect in numbers is what separates one portfolio from another. "Works with high efficiency" means nothing, whereas "raised classification accuracy on my test data from 70% to 90% against a simple baseline" (an invented example of phrasing, not a real result) lets the reader judge it.

Document as practitioners do

The paper "Model Cards for Model Reporting", by Margaret Mitchell and colleagues (2018), proposed that every trained model should come with a short document describing its intended uses, how it was evaluated and how it performs across different groups. You can borrow the idea for each project, writing in a clear readme:

• The problem, and who it matters to.
• The data: its source, its licence and what was excluded.
• The method, and why you chose it over the alternatives.
• The results, how you measured them and how they compare with a baseline.
• The limits: where the model errs, who it does not suit and what you did not test.
• How others can rerun the work.

Documentation is part of the work, not an ornament added afterwards; it shows your ability to communicate and your integrity in stating limits.

Honesty in presentation

A portfolio is worth only as much as it is truthful. Credit everything you took from others, respect the licences of data and code, do not publish data about real people without their consent, and do not inflate a result or hide a failure. Showing a mistake you made and how you corrected it adds to your credibility rather than subtracting from it, for honesty and trustworthiness are among our values in knowledge and in work.

In the Saudi context

In October 2020 the Saudi Data and Artificial Intelligence Authority (SDAIA) announced the National Strategy for Data and AI, whose stated aims included training 20,000 data and AI specialists and seeing 300 active startups established in the field. These are declared targets, not achieved results. Arabic offers a wide field for locally rooted projects (see the article "Arabic Artificial Intelligence: Do We Make or Consume? Sovereignty and Arabic Language Models"): classifying texts, summarising documents, or a tool to serve a small charity.

So begin with a small project you can finish in a few weeks, document it, and build the second on what you learned from the first. A portfolio is a habit practised over time, not a file assembled just before an interview, and success is from God.