LLM Researcher at Multiverse Computing. I train, compress, and optimize large-scale foundation models for both language and vision.
PhD in Natural Language Processing from the University of the Basque Country UPV/EHU, developed at the IXA Group and the HiTZ Basque Center for Language Technologies.
I have strong, hands-on experience in performance optimization and HPC across academia and industry. At HiTZ I trained Latxa on the Leonardo Supercomputer using up to 512 GPUs across 128 nodes. At Krea I optimized fine-tuning pipelines for image and video diffusion models with distributed techniques such as FSDP2. At Multiverse Computing I optimize inference and fine-tuning pipelines, achieving 3× faster logit precomputation and 5× faster KL-divergence training.
Exponential Fellow 2025. In my free time I build veridika.ai, an AI agent framework for real-time fact-checking.
Featured Models
