Data Engineer with 4 years in high-volume financial systems at BBVA Technology (DB2/SQL, business-critical batch processing) and hands-on experience building ETL pipelines with Azure Data Factory, SQL Server and Python.

MSc in Data Science, Big Data & Business Analytics — Universidad Complutense de Madrid, 2026. First prize in the master's scholarship competition for the thesis below.

Interested in data architecture, distributed processing, cloud solutions and GenAI-powered data workflows.

🟢 Available for new opportunities — Madrid (hybrid) or remote.

MSc thesis (UCM) · First prize, master's scholarship competition · Team of 3 — my part: infrastructure, 24/7 capture and the web app

Renfe reports commuter rail delays only once they are already happening, and publishes no history. There was no raw material to train anything, so I built the system that creates it.

• 12 million records in 92 days, captured every minute from three RENFE GTFS-Realtime feeds, hourly AEMET weather and the daily official timetable
• 99.7 % measured availability with no manual intervention, on a 5 €/month Linux VPS
• Raw→Parquet medallion architecture, nightly compaction, cloud backup, and content monitoring that alerts within 10 minutes when a source stops making sense
• Lifted the real-time × official timetable join from 0 % to 99.9 % by recovering historical timetable versions the operator does not keep. Without that fix there was no trainable dataset
• Project outcome: LightGBM model cutting error from 283.2 s to 147.7 s (−48 %) vs the historical baseline. On real trains, 6.18 min error vs 9.08 for the official timetable
• Public web app with arrival prediction, incidents, live map, two languages and a conversational assistant that cannot invent figures — the LLM only parses the question, the system composes the answer from real data

Python · pandas · Parquet · Linux · systemd · cron · LightGBM · FastAPI · rclone

🔗 Open the live app

End-to-end ML pipeline for DrivenData's Pump It Up over 59,000 water pumps in Tanzania: stacking ensemble scoring 0.8230, imbalanced-class strategies for a critical class representing only 7.3 % of cases, 2× detection rate of repairable pumps vs base models. Includes an LLM-based EDA agent and a conversational "Ask the Project" assistant.

scikit-learn · XGBoost / LightGBM · LangGraph · Anthropic API · Gradio · SHAP

Business-oriented SQL analytics on revenue, profitability and commercial performance.

SQL

In production — used professionally at Babel and BBVA Technology

In my own projects — built and operated by me, see above

Academic — MSc coursework, not production experience

Azure and Databricks certifications in progress.

• MSc in Data Science, Big Data & Business Analytics — Universidad Complutense de Madrid, 2025–2026
• BSc in Industrial Mechanical Engineering — Universidad de Huelva, 2011–2017
• Algoritmos de Big Data para Ingenierías — Global Knowledge, 2026 (80 h: data mining, R, feature engineering, supervised modelling)
• Snowflake University Platform Skills Badge — 2026