Mathis Dumont
About
I'm a Data Scientist intern at Mirakl, where I fine-tune open-weight models and align them with DPO to automate catalog onboarding. Before that, I built a multi-agent LLM pipeline at ArcelorMittal to reverse-engineer 200k+ lines of legacy Fortran.
I also have a commercial background. From 2023 to 2025 I led sales at a Junior Enterprise turning over €250k a year, and we delivered projects for more than 50 companies.
Skills
- LLMs:DPO • Fine-tuning • LLM-as-a-Judge • Multi-agent systems
- Data Engineering:PySpark • Databricks • Delta Lake • SQL
- Machine Learning:PyTorch • Hugging Face • Scikit-learn
- Infrastructure:Docker • FastAPI • AWS (S3) • CI/CD (GitHub Actions)
Awards
- Finalist, Gradient 2026:national coding competition run by Prologin & Mistral AI.
- 2nd Prize, 12th Trinational Students' Congress of Natural Sciences & Technology, Freiburg:IoT smart beehive.
- 1st Junior Enterprise in France, 2024:Arts & Métiers Junior Enterprise, where I was Head of Sales.