ML researcher & engineer working on multimodal learning, foundation models, LLMs and biomedical AI. M.S. Computer Science (AI) at UC San Diego · Fulbright Scholar · B.S. Biomedical Engineering & B.Eng. Telecommunications/Electronics Engineering (University of Barcelona).
Machine Learning Researcher — Swartz Center for Computational Neuroscience, UC San Diego (Jun 2026 – present) Multimodal physiological and behavioral signals: EEG, eye-tracking, motion capture, electrodermal activity, synchronized at sub-millisecond precision.
- 📄 Designed and led a large-scale multimodal ML dataset & benchmark unifying eight sensor modalities. First-author paper submitted to ICLR 2027; under double-blind review, so code, data and paper will be released after the review period.
- Fine-tuning transformer-based EEG and time-series foundation models against self-supervised and classical baselines; building late-fusion multimodal models across sensor streams
- Engineering reproducible pipelines that turn tens of GB of raw sensor recordings into validated, ML-ready datasets, scaled on multi-GPU SLURM clusters
- Designing benchmark evaluation methodology and representation/transfer-learning baselines
NLP / ML Engineer — Barcelona Supercomputing Center, NLP4BIA (Jul 2023 – Aug 2025) Biomedical & clinical NLP for Spanish-language health data.
- Author and lead developer of KeyCARE: open-source library for biomedical keyword extraction, categorization and relation extraction (upstream)
- Fine-tuned Transformers and LLMs (instruction tuning, prompting, RAG) for classification, entity/relation extraction and normalization
- Contributed to the design of TriAgent Pediatrics: a multi-agent LLM + guideline-rule system for pediatric triage (team design of the agent architecture, triage rules and evaluation)
- Co-led technical development of TeresIA, a multi-institution Spanish government AI initiative for terminology and information retrieval
| Area | Project | What it is |
|---|---|---|
| 🧠 LLMs · RAG | Hybrid-RAG-QA | RAG QA system: hybrid FAISS + BM25 retrieval, reranking, multi-query, LLM-judge evaluation |
| 🧠 LLMs · Fine-tuning | LLM-Schema-Linking-Text-to-SQL | LoRA fine-tuning of Qwen2.5 for text-to-SQL schema linking with large experiment grids |
| 🧠 Pretraining | NanoGPT-Pretraining-100M | GPT-style language model under a 100M-parameter budget, with scaling & optimizer ablations |
| 🧠 Deep learning | Transformer-Encoder-Decoder-From-Scratch | Transformer encoder & decoder implemented from scratch in PyTorch |
| 🏥 Clinical NLP | KeyCARE | Python library (PyPI) for biomedical keyword extraction, categorization and relation extraction |
| 🏥 Medical imaging | SISCOM-Epilepsy-Localization | SPECT/MRI pipeline for localizing the epileptogenic zone |
| 📈 Signals | ECG-Signal-Classification | From-scratch Pan-Tompkins R-peak detection + HRV features for atrial-fibrillation classification |
| 🧬 Genomics | IBD-Relative-Finding-Benchmark | Benchmark of IBD detection methods for relative finding on 1000 Genomes |
| 📊 Modeling | COVID19-SIR-SEIR-Modeling | Epidemic ODE models (SIR/SEIR + quarantine) fitted to real COVID-19 data |
| 📊 Classical ML | 1NN-Prototype-Selection | Prototype selection strategies for 1-NN classification on MNIST |
| 🤖 Robotics | ROS2-EKF-SLAM-Path-Planning | ROS 2 robot: EKF-SLAM, roadmap path planning, coverage navigation |
| 🔌 Embedded | MSP430-Robot-Control-PCB | Bare-metal C firmware + custom KiCad PCB for a robot |
ML: PyTorch · Hugging Face (Transformers, PEFT) · scikit-learn · sentence-transformers · FAISS · spaCy LLM systems: RAG · LoRA / instruction tuning · multi-agent workflows · LLM-as-judge evaluation Infra: SLURM / multi-GPU · Docker · AWS · Flask · Git Languages: Python · SQL · C/C++ · MATLAB · R · Java
Fulbright Scholarship (2024) · Extraordinary Degree Award, B.S. Biomedical Engineering — Rank 1 (2024) · First Prize, Gemma Rossell Thesis Awards (2024)