Data and artificial intelligence internship: start from a useful question

What a data or AI internship really covers

Behind the words "data" and "artificial intelligence" sit neighbouring but distinct jobs. The data analyst collects, cleans and interprets figures to inform a decision. The data scientist builds statistical or machine learning models to predict or classify. The machine learning engineer puts those models into production and monitors how they behave. The data engineer, finally, builds the pipelines that carry clean data to the people and systems that depend on it.

An internship rarely covers the whole chain. It is usually one link: preparing a dataset, testing a modelling idea, building a dashboard, or labelling examples to train a system. Knowing which link appeals to you lets you write a targeted application rather than a generic message sent to every "data" team in a region.

Host organisations vary widely: departments of large companies, software vendors, consultancies, research labs, hospitals, local authorities, and charities with a statistics unit. Each offers a different pace and level of supervision, which matters as much as the title of the advert.

The tasks an intern is actually given

In the first weeks the work often looks like detective work. You are handed files from several systems, sometimes incomplete, and you must work out what each column represents, spot outliers and document your cleaning choices. This unglamorous work commonly takes up half the time of a data project.

More visible tasks follow:

• producing indicators and charts that non-specialists can read;
• comparing several methods on the same problem, starting with the simplest;
• testing a classification, forecasting or language-processing model on a limited scope;
• labelling or quality-checking the examples used to train a system;
• writing a note that explains the method, its limits and the errors observed.

That last task is underrated. A result that nobody can re-read or reproduce has little value for the team. Showing at interview that you can write and explain sets you apart from candidates who only present scores.

Useful skills: a technical base and working habits

The technical base comes down to a few points. You need a programming language commonly used in analysis, most often Python and sometimes R, plus SQL to query databases. Then come descriptive and inferential statistics, reading a confusion matrix, the difference between correlation and causation, and an understanding of the risk of overfitting. Computing and visualisation libraries are quickly learned if these foundations are solid.

For AI proper, some knowledge of neural networks, language models or computer vision is appreciated, but a supervisor often prefers a candidate who masters classical methods and can say when they are enough.

Working habits weigh at least as much:

• rigour, notably in keeping training data and test data separate;
• curiosity about the business, because understanding the field avoids answering the wrong question;
• clarity in speech and in writing;
• a critical eye for any result that looks too good to be true.

Building a portfolio project that holds up

A convincing project starts from a precise question, not from a dataset chosen because it is famous. For example: "Which factors explain delays in a delivery service?" or "Can we automatically spot customer reviews that report a safety defect?" The question defines what will be measured and what counts as success.

Here is a simple structure, useful for a public repository or a PDF file:

1. the question and why it matters to a real user;
2. where the data comes from, its licence and its known gaps;
3. the cleaning carried out and the rows set aside, with the reasoning;
4. a very simple baseline, such as an average or a manual rule;
5. the model chosen and the reason for choosing it;
6. the errors observed, with a few commented examples;
7. the limits and what should be done next.

Open data portals let you work on real, documented sources. Just make sure you read the reuse licence before publishing your results.

Personal data, ethics and the legal framework of the internship

Working with data comes with rules. As soon as a file concerns identifiable people, the European data protection regulation applies: a clear purpose, data minimisation, a retention period and the rights of individuals. A serious organisation will tell you which data you may handle, from which workstation, and what you may or may not take with you at the end of the internship. Ask in the first days; doing so shows professionalism.

AI systems are also covered by a specific European framework, being phased in gradually. For an intern, the practical point is to find out whether the tool touches sensitive decisions (recruitment, credit, health) and to understand that documentation and human oversight requirements then apply.

On the contractual side, in France an internship rests on a three-party agreement signed before the first day by you, the host organisation and your school or university. It sets out the tasks, the duration, any stipend, the supervisor and the confidentiality clauses. Read it carefully: those clauses decide what you will be able to show afterwards in a portfolio.

Duration, pay and a worked example for illustration

The length of a data internship depends on the course: from a few weeks at bachelor level to four or six months at the end of a master's or engineering school. In France, beyond two months of presence in the same organisation, a stipend (gratification) is compulsory. Its minimum is set by regulation as a percentage of an hourly Social Security ceiling, and it changes over time: check with your school or the host organisation, which may also pay more than the minimum.

Illustration with round, hypothetical figures. Imagine a five-month internship with a stipend of 1,000 euros a month. The total received would be 5,000 euros, before any deduction for transport costs. If the same organisation offered 1,200 euros, the difference over the period would come to 1,000 euros, but it should be weighed against the quality of supervision, the distance from home and the cost of housing, which can matter more than the gap in pay.

In public research, amounts depend on the rules of each body; ask it directly before comparing offers of a different nature.

Applying: CV, cover letter and technical interview

On a CV for a data internship, a line reading "Python, SQL, machine learning" is not enough. Tie each tool to something you achieved: "cleaned 40,000 sales rows and built a seasonality indicator" says more than any list. Add the link to your code repository only if it is readable, with a clear introductory file.

A cover letter works better when it names a problem faced by the target organisation: stock management, patient follow-up, a flow of documents. Show that you have looked at its publications, its website and its tools. Avoid phrases about a "passion for artificial intelligence", which tell the recruiter nothing.

At interview, expect three kinds of questions:

• a detailed presentation of one of your projects, including the choices and the failures;
• a small exercise, such as commenting on a chart or writing an SQL query;
• common-sense questions: how would you know your model is wrong? What would you do if data were missing?

Being able to answer "I don't know, but here is how I would find out" goes down better than an invented answer.

When and where to look, and how to choose between an internship and a work-study contract

End-of-studies internship offers are often published from autumn to early spring, for starts in spring or summer. Research bodies and large companies sometimes ask for applications several months ahead. Small teams recruit later and faster. A well-targeted speculative application remains a useful route, especially to the statistics units of local authorities or charities.

Schools and universities have placement offices and alumni networks. Their own job boards and graduate job sites let you filter by keywords such as "data", "analysis" or "machine learning".

As for choosing between an internship and a work-study contract (alternance), it depends on the rhythm you want. In France an internship is a training arrangement governed by an agreement, with student status. Work-study is a paid employment contract lasting one to three years, alternating between school and company. It gives more continuity on one data topic, but you must secure a host company before the academic year starts.

Common mistakes to avoid

Several slips come up again and again among candidates at the start of their career.

• Chasing the most complex model without comparing it to a simple solution. Teams want to know what the model adds.
• Ignoring where the data comes from: a dataset published without a description cannot support a reliable conclusion.
• Data leakage: letting into training any information that will not be available in real use distorts the whole evaluation.
• Presenting a single performance figure without showing the errors, the rare cases or the groups that are handled badly.
• Reusing the host organisation's data in a portfolio without written permission.

Another trap is believing the internship will only be about learning a tool. Supervisors mostly judge the ability to ask questions, to reformulate a vague need and to deliver, even modestly, something usable by the agreed date.