Research

My research lies at the intersection of knowledge graphs, LLMs & NLP, graph learning, and recommender systems.
A common thread is the use of structured information and learning on graphs to solve real-world problems that require both performance and explainability.

I work with academic and industrial partners on applications in recruitment, medical decision support, and financial markets.

Research themes

Knowledge graphs

I work on building knowledge graphs and using them in machine learning problems.
I am particularly interested in how to extract, clean, and organize knowledge from heterogeneous sources, and how to make this knowledge usable in downstream AI systems.

LLM & NLP

I use NLP tools and LLMs to build and enrich knowledge graphs, and I explore hybrid approaches that combine LLMs with graph learning.
This includes using language models to extract structured knowledge, reason over it, and interact with users, while keeping a strong link to explicit graph representations.

Graph learning

I do machine learning on graphs (especially on knowledge graphs) to solve real-life problems that require both high performance and explainability.
My work spans from designing graph representations to developing models that can leverage structure, time, and semantics.

Recommender systems

I build ranking and recommendation systems, often using graph-based models.
A current focus is on recommendation scenarios where data is heterogeneous, temporal, and sparse (e.g. cold start, evolving user profiles).

Selected projects

Job recommendation with heterogeneous data

I led a collaboration with a company, with whom I supervised a CIFRE PhD student.
We built a recommender system using a temporal heterogeneous graph and LLMs to assist professional recruiters.

• Job recommendation on a heterogeneous, temporal graph
• Integration of LLM-based representations for profiles and job offers
• Focus on cold start, recruiter workflows, and practical deployment

Related publications (selection):

• Tackling cold start for job recommendation with heterogeneous graphs
• TIMBRE: Efficient job recommendation on heterogeneous graphs for professional recruiters

You can find the full references on the Publications page.

Temporality in medical data

I lead a collaboration with a medical start-up that builds tools to assist doctors, with a CIFRE PhD student.
We study how to integrate temporal medical data into reasoning with LLMs to better ingest and summarize relevant information from a patient’s medical history.

• Temporal modelling of structured and unstructured medical data
• Reasoning with LLMs over patient histories
• Decision support for medical professionals

Publications are in preparation for this project.

Explainable financial time series prediction from exogenous sources

This is a research project I initiated, funded by an IMT Futur & Ruptures scholarship, with one PhD student.
We study how to reason jointly on time series and news articles to better understand perturbations in financial markets.

• Joint modelling of financial time series and textual news
• Explainable predictions and attribution of market moves to events
• Combination of temporal models, NLP, and structured reasoning

Publications are in preparation for this project.

Research philosophy

A few principles guide the way I do research:

• Ethical research. I aim for research that benefits society, avoids shortcuts and questionable practices, and is honest about limitations. This includes trying to critique even our own results and making space for negative results when they are informative.
• Open research. Whenever possible, I release code, papers, data, and even reviews. I have signed the No free view? No review! pledge (see https://nofreeviewnoreview.org/).
• Quality over quantity. In a context where the number of AI submissions and publications is exploding, I believe we should resist the pressure to submit unfinished, non-reproducible, or low-impact work just to increase publication counts.

These values strongly influence how I supervise students, choose projects, and collaborate.

Collaborations

I am open to many forms of collaboration, primarily around research:

• joint research projects (academic or industrial),
• PhD and master’s supervision,
• applied research collaborations with companies or public institutions,
• consulting on AI, knowledge graphs, LLMs, and graph-based recommender systems.

If you are interested in working together, feel free to contact me by email.