Research Engineer · CENIA

Tomás Couso Coddou

I am a Computer Science Engineer from Pontificia Universidad Católica de Chile working on machine learning, with a focus on reinforcement learning problems. I am currently a Research Engineer at CENIA, within its Reinforcement Learning and Robotics team. My trajectory began in neuroscience and psychology, influenced by Francisco Varela’s framework of embodied cognition. Transitioning to engineering allowed me to tackle these same systemic questions from a quantitative, computational perspective.

After finishing my degree, I set out to gain engineering experience with large-scale foundation models, and joined the team behind LatamGPT: a model trained to reflect the languages, expressions and contexts of Latin America rather than inherit the biases of imported systems, developed through more than 50 technical alliances across 15 countries. It launched in February 2026 at a public ceremony led by the President of Chile. Building it gave me hands-on engineering experience with large-scale data pipelines and distributed training regimes.

Currently, my focus is on latent world models and trajectory modelling for reinforcement learning problems. Our recent work at CENIA involves diffusion-based policies for behavioral cloning and JEPA architectures with action encoding for Model Predictive Control (MPC), where learned action representations make planning stable over long horizons.

Location

Santiago, Chile

Interests

World Models Diffusion Policy JEPA Reinforcement Learning

Minimax search was one of the first algorithms I ever studied and implemented — here's the code. You play as X.

01Experience

Centro Nacional de Inteligencia Artificial Apr 2026 – Present

Research Engineer · Santiago, Chile

Working on world models for robotic control: diffusion-based policies for behavioural cloning and JEPA architectures with action encoding.

Centro Nacional de Inteligencia Artificial Apr 2025 – Mar 2026

ML Engineer · Santiago, Chile

LatamGPT: Latin America's largest open multilingual LLM, launched to wide public recognition. Owned the end-to-end data ingestion pipeline, from tokenization and padding through dataset assembly and curriculum structuring, using PySpark and Kubernetes. Authored the corresponding sections of the technical report (sections 2.1.2 and 2.1.3). Also ran pre-training data distribution experiments.

National Institute of Informatics Dec 2024 – Apr 2025

Research Intern · Tokyo, Japan

Competitively selected for the 2024 NII International Internship Program under Prof. Yusheng Ji. Developed reinforcement learning methods for UAV-based agricultural territory monitoring, leading to a publication at ICRA 2026.

Pontificia Universidad Católica de Chile Mar 2023 – Dec 2024

Research Assistant · Santiago, Chile

Built a deforestation detection system using satellite remote sensing imagery and Echo State Networks. Published at a NeurIPS 2024 workshop.

Lemu Dec 2023 – Feb 2024

ML Engineering Intern · Santiago, Chile

Built a satellite imagery data pipeline for a land cover classification algorithm at a startup focused on biodiversity monitoring and conservation.