Carlos Rosas-Hinostroza

Data Scientist, LLM Training, Digital Sociology. Based in Paris.

About

Senior Data Scientist at Pleias, specializing in developing AI solutions with a commitment to open science principles.

PhD candidate at École Normale Supérieure (LATTICE), a CNRS laboratory.

Sociologist, mainly interested in the study of online cultural phenomena.

Former lawyer with experience in trademark law.

Current Positions

Senior Data Scientist — Pleias

2024 — Present

Corpus collecting and curation for the training of language models and other AI solutions under an open science philosophy: open weights, open corpus and open code. Projects developed for RATP, Wikimedia, Libraries Without Borders and many others.

Lecturer — Sorbonne Nouvelle

2025 — Present

In charge of the Data Visualization courses for the Digital Communication and Data Analysis Master.

Education

PhD · NLP and Social Sciences, École Normale Supérieure – LATTICE (2025 — Present)
MA · Contemporary Sociology, Sorbonne Université (2024)
MSc · Data Science & Digital Society, Université Gustave Eiffel (2023)
BA · Sociology, Sorbonne Université (2020)
Law Degree, Universidad Nacional Mayor de San Marcos (2014)

Selected Works

Common Corpus — The largest open licensed text corpus in the world (2.27 trillion tokens).
French Science Commons — 1.2 million French scientific publications, curated and transformed to markdown format.
On Device RAG — A 0.3B mini language model for RAG with reasoning, analysis of sources and full references quoting, capable of running on a Raspberry Pi.

Publications

Common Corpus: The Largest Collection of Ethical Data for LLM Pre-Training. ICLR 2026 (Oral), 2026.
Model in Distress: Sentiment Analysis on French Synthetic Social Media. ACL 2026, 2026.
Pleias 1.0: The First Family of LLMs Trained on Fully Open Data. Procedia Computer Science, 2025.
Even Small Reasoners Should Quote Their Sources: The Pleias-RAG Family. Preprint, 2025.
Old Speeches, New Resources: Memes as a Far-Right narrative tool in Peru. Letras, 2024.