Professional pianist (classical and flamenco) transitioning into data science and AI. Piano performance and programming share a common foundation: analytical rigour, pattern recognition, and discipline in complex structures. I hold a Título Superior in Piano Performance (GPA 9/10), a Master's in Advanced Artistic Studies in Flamenco — Flamenco Piano from ESMUC (GPA 9.5/10), and a Master's in Neuromusic (UCAM, 60 ECTS, título propio) — genuine domain expertise in audio and music that I bring to technical work where that background actually adds value.

Currently completing the Data Science Career Accelerator — University of Cambridge Professional and Continuing Education (PACE), an intensive, project-based professional certificate (non-credit), alongside strong Python fundamentals from Escuela Musk.

My current focus is Music Information Retrieval: applying machine learning to flamenco audio, with a view to doctoral research in computational musicology.

• Python — advanced concepts, OOP, modules, file handling, JSON/CSV, automation
• Data Science & ML — scikit-learn pipelines, clustering (K-Means, hierarchical), dimensionality reduction (PCA, t-SNE), classification, model evaluation
• Web & APIs — Flask, Django fundamentals, REST APIs
• DevOps — Git/GitHub, Linux, Docker, CI/CD (GitHub Actions)
• Audio & Music (specialisation) — librosa, audio feature extraction (MFCCs, chroma, spectral features), music theory applied to ML

• 🎸 flamenco-classifier — automatic classification of flamenco palos (soleares, seguiriyas, bulerías, alegrías) from audio. Features are grounded in music theory rather than picked blind — Phrygian mode and the Andalusian cadence, position of the metric close within the 12-beat compás — and fed to a Random Forest: cross-validated F1-macro 0.645. Live demo — upload a track and get the predicted palo with its probabilities.
• 🎤 cante-embeddings — embeddings of a cantaor's interpretive identity (who is singing, not which palo), built by fine-tuning pretrained OpenL3 features with a triplet-loss projection head. Closed-set pairwise verification: AUC 0.80, with the small-sample limitations documented rather than glossed over.
• 🔍 Customer Segmentation with Clustering — course mini-project: unsupervised ML on a 951K-row e-commerce dataset; K-Means + hierarchical clustering, PCA/t-SNE, translated into a marketing strategy.

📌 See all pinned repositories below for the full picture.