Data science · ML · Madrid
Data science, learned through projects
I study data science and machine learning in Madrid, and I practise it with complete projects: from exploratory analysis and problem definition to a served model. Tabular data and text, trying to keep the methodology solid and the evaluation honest at every step.
The analysis that precedes the model
Before training anything, I analyse data distributions, missing values, correlations with the target variable and potential biases. I learned the hard way that a model trained on poorly understood data does not work even if the algorithm is sophisticated.
Feature engineering that contributes
The features you build from raw data usually matter more than the algorithm. I practise feature engineering carefully: ratios, temporal aggregations and encodings that do not leak information from the future into the past.
Models with methodology
Stratified cross-validation, hyperparameter search with a fixed budget, model comparison on the same test set, probability calibration and feature importance analysis. No shortcuts that inflate results, which is the easiest trap when nobody is checking your work.
Results you can actually use
Notebooks are for exploring, not for delivering. I try to make every project end in a reproducible pipeline, an API or a clear README with the conclusions. Not a Jupyter notebook that only runs on my machine.
Feedback, an idea or an opportunity?
The projects are on GitHub with their evaluation and documented decisions. If you want to tell me something, write to me.
Write to meFAQ
What is the difference between data science and machine learning?
Data science is broader: it includes exploration, statistical analysis, visualisation and narrative about data. ML is a tool within data science, the part about building predictive models. In practice they go together in almost all my projects.
What types of data do you practise with?
Mainly tabular data (credit risk, customer churn) and text (classification, extraction, summarisation, semantic search). I use public datasets and synthetic data documented in each repository, so anyone can reproduce the results.
How do you choose which project to build?
By what I want to learn. If I read about probability calibration, I look for a problem where the probability matters (credit risk). If I want to understand transformers, I implement one from scratch. The project is the excuse to land the theory.
Where are the results?
Every project is at github.com/delcenjo with its code, metrics and documented decisions. The more general lessons I write up on this site's blog: data leakage, honest evaluation, when an LLM is worth it.
Can I write to you?
Yes, please. Feedback on the methodology, project ideas, questions, or an internship or first junior role in data. The contact form goes straight to my inbox.