Projects
Selected works across Applied Artificial Intelligence, Deep Learning, Supercomputing, and Full-Stack Engineering.
Artificial Intelligence for Cognition — Decoding Cognitive Conflict & Perception from EEG
Bachelor's thesis evaluating Riemannian geometry, Spatio-Temporal GNNs, LoRA-adapted Transformers, and EEG Foundation Models to decode cognitive conflict during binocular rivalry.
Vegan Tools
A production-grade web platform featuring an Ingredient Checker, Menu OCR Interpreter, and Recipe Veganizer built with Next.js and multi-modal AI.
Large-Scale Transformer Training on MareNostrum 5
Distributed Vision Transformer training on the 200 GB+ MAMe dataset using the MareNostrum 5 supercomputer at BSC.
SinergIA — Multi-Agent Systems & LLM LoRA Fine-Tuning
Multi-agent conversational architecture with LangGraph and Google Gemini, fine-tuning clinical experts with LoRA to boost MedQA accuracy.
IEEE BigData 2024 Cup — Predicting Chess Puzzle Difficulty
Ensemble machine learning model predicting human Elo difficulty of tactical chess puzzles leveraging Stockfish, Leela Chess Zero, and Maia.
In-Context Learning in LLMs for Catalan NLP
Empirical research paper evaluating in-context learning, few-shot prompting, and reasoning capabilities of open-weight LLMs on Catalan text manipulation tasks.
Large-Scale Data Engineering & Knowledge Graphs
Architected distributed data pipelines integrating heterogeneous datasources into semantic knowledge graphs with GraphDB and high-throughput SPARQL queries.