ML Engineer · Computer Vision & LLMs
I take vision and language models from research to production: industrial defect detection and segmentation, LoRA-tuned multimodal OCR and RAG systems. Trained as a physicist and mathematician, I care about understanding why a model works before shipping it.
Elche, Alicante, Spain
About me
Physicist and mathematician specialized in Machine Learning and Computer Vision engineering. I take vision and language models to real production: from data to deployment with Docker, vLLM and MLflow. I care about understanding models deeply, not just training them. When I am not training models, I write about physics, chaos and meteorology on this blog.
...
Experience
1
Publications
5
Open source
A ~2 KB differentiable filter that undoes an illumination shift before the image reaches a frozen detector. No retraining, no labels, minutes on a CPU.
An LLM + RAG architecture that builds a verified ontology incrementally and uses it to flag false claims in films, podcasts and texts. Published at DCAI'25.
Defect detection and segmentation (CNNs, ViT, YOLO, RF-DETR) plus sidewall reading with a LoRA-tuned GLM-OCR, in production at VRAIA. Served with vLLM and Docker.
I take defect detection and segmentation models (CNNs, ViT, YOLO, RF-DETR) and a multimodal OCR (GLM-OCR) to production for industrial tire inspection: mAP@50:95 of 0.83 and 98.56% exact match reading sidewalls.
I write about this work in OCR in extreme conditions and CNNs vs Vision Transformers.
Research in NLP and LLMs: an ontology-based misinformation-detection architecture built with LLM-generated ontologies (published at DCAI'25), RAG pipelines and multi-agent systems, plus computer vision projects.
Publication: DCAI'25 paper.
I developed Lagrangian methods to automatically identify atmospheric blocking situations — the basis of my thesis and a publication in progress.
I tell the full story in Atmospheric blocking, a Lagrangian view.
Cómo construimos, en un trabajo publicado en DCAI'25, un sistema que genera automáticamente una base de conocimiento (ontología) con LLMs y RAG, y la usa para detectar información falsa en películas, podcasts y otros formatos de ficción.
Qué pasa cuando llevas el OCR a un entorno industrial real —texto en relieve negro sobre negro, polvo, brillos y cadencia de línea— y qué aprendí probando GLM-OCR, un modelo multimodal de 0.9B parámetros que corre en local
Un recorrido por la interpretabilidad en visión por computador: filtros clásicos, mapas de características, Grad-CAM y atención. Y por qué visualizar lo que ve tu modelo no es opcional en producción
Universidad de Alicante
+120 ECTS from the Mathematics degree taken in parallel (UA simultaneous-studies program).
Thesis: A Lagrangian method for the identification of atmospheric blocking situations
Universidad de Salamanca
International mobility scholarship.
AWS Cloud Technical Essentials
Amazon Web Services
Build Basic Generative Adversarial Networks (GANs)
DeepLearning.AI
Introduction to Front-End Development
Meta
Cambridge English B1 Preliminary
Cambridge Assessment English
AI-Powered Ontology-Based Architecture for Misinformation Detection in Fiction Works
Marcos Arias-González, Juan Manuel Ruiz Muñoz, Pablo Armenteros Cosme, Lucía Isabel Rodríguez González, Javier Curto Hernández
BISITE Research Group, University of Salamanca · TRUESTORIES project (CPP2021-008358)
A Lagrangian Method for the Identification of Atmospheric Blocking Situations
Got a computer vision or LLM problem to take to production? Drop me a line.
