AI / Machine Learning Engineer · Computer Vision
2 years in ML: vision, NLP, tabular · MSc in AI (UPV) · Spain
2 years in ML: vision, NLP, tabular · MSc in AI (UPV) · Spain
 viola-jones-face-detection Haar features, AdaBoost cascade, hard-negative mining and a real-time Numba webcam detector. Built from scratch, no pretrained detector    
|
 multimodal-sexism-detection-tiktok Multimodal classifier (text, biosignals, video) for the EXIST 2026 challenge. 80.6 macro F1 on the binary task (project presentation)  
|
 glioblastoma-survival-missing-modalities Survival classification from 3D MRI, radiomics and clinical data, built for missing inputs. Best test F1 0.685 (single runs, 61-patient test set)  
|
 voice-assistant-whisper-rag-tts Speech-to-speech RAG assistant; local and cloud pipelines compared. 60 tests (30 local, 30 cloud); best mix about 7.5 s per answer    
|
 multi-agent-expert-debate LLM experts debate until a reviewer confirms consensus; a moderator writes the decision. Up to 5 debate rounds, 12 ready-made debates   
|
 crowd-evacuation-simulation Agent-based evacuation: density, behaviours, exits and pathfinding. 1000 runs per configuration across 7 dimensions 
|
Several of these were built together with Miquel Gómez Corral; each repository says which. Diagrams for the Viola-Jones, sexism-detection and debate cards are original summaries of each repository's README.
| Languages |    |
| ML / DL |         |
| LLM and RAG |     |
| Deployment |      |
| Optimisation |     |
AWS, Azure and CI/CD: working knowledge.