Applied Tabular AI Intern

About Probabl

Probabl is the Tabular AI company founded by the creators and stewards of scikit-learn. We build Skore, the Tabular AI platform whose agent builds, validates and tracks predictive models with the statistical methodology of the scikit-learn maintainers; we steward scikit-learn and its ecosystem (skrub, joblib, imbalanced-learn, fairlearn, skops, hazardous); and we contribute to TabICL, the open source tabular foundation model. Our customers are large groups, often in regulated settings, who need reliable predictions, not just faster ones.

The internship

You join Probabl's Services team, the applied AI team between our clients and the engineers who build the product, for six months in Paris. The team runs the demos, proofs of concept, workshops and hackathons through which banks, insurers, energy and industrial groups see what reliable tabular AI looks like on their own data.

This is not an observation internship. You work with the Solution Architects and the Forward Deployed Engineers on live engagements, on real client data rather than toy datasets, with a senior engineer alongside you at every step.

What you will do

• Help prepare and co-run client sessions: demos adapted to a prospect's sector and data, workshops, hackathons and proof-of-concept working sessions. You present in them, with a senior engineer next to you, not in front of you.

• Build proofs of concept on client data: scoping with the architect, feature engineering with skrub, baselines with scikit-learn, evaluation and leakage detection with Skore, and a readout the client can act on.

• Run engagements agent-first with the Skore agent and LLM coding agents, find where they are fast and where they go wrong, and turn that into reusable skills and checks for the team.

• Close the loop with product and open source: reproduce field issues and open them on Skore, skrub or scikit-learn.

What you will learn

Tabular ML at the level of the people who wrote the library, how enterprise AI actually gets adopted, what it takes to be trusted by a client, agentic data science in practice, and open-source contributions reviewed by their maintainers. The Services team is growing, and a successful internship is the natural way into a full-time Forward Deployed Engineer, Applied AI or Solutions Engineer role.

Who we are looking for

• A final-year or gap-year student at a leading engineering school or research master's.

• Strong scientific Python and end-to-end tabular projects behind you (competitions, research, side projects), not only coursework.

• You care about evaluation: cross-validation done right, leakage, calibration, business metrics, and saying "this does not work" when it does not.

• You can explain a model to a non-technical stakeholder, in French and in English, and you are comfortable in front of a client.

• Curious about tabular foundation models and LLM agents for data science. Daily use of a coding agent, open-source contributions or a deployment you have done yourself are a plus.

Practical details

• Team: Services (applied AI and field engineering), reporting to the VP Services, co-founder of Probabl.

• Location: Paris.

• Duration: 6 months, final-year internship or gap year. Start: flexible.

• Compensation: monthly internship stipend (gratification de stage), to be determined.

• Languages: English.

• Requirements: eligible for a tripartite internship agreement (convention de stage) and authorized to work full-time in France for the entire duration of the 6-month placement. Visa sponsorship is not available.

How to apply

Send your CV, a link to something we can read (GitHub, a project, a competition result), and five to ten lines on the most interesting tabular data problem you have worked on and what you learned. No cover letter.

The process takes about two weeks: a 30-minute call with the VP Services, a short technical case on a tabular dataset (about one hour, live or take-home), and a conversation with an FDE or a Solution Architect.