Tools & technologies
Tabular data modelling
Most real problems are tabular data.
Most real problems are tabular data. Models to predict default, churn or risk, with the evaluation an actual decision needs.
Calibration, not just ranking
If you are going to make decisions with the probability, it has to be reliable. I calibrate the model and check it with the Brier score and the reliability curve, so that a 0.7 really means 70%.
The threshold matters more than the model
I pick the decision threshold that minimises the real business cost, not the default 0.5. That often changes the outcome more than changing the model.
Operating locations
Studying in Madrid
Based in Madrid, learning in public: projects, open-source tools and collaborations, inside and outside Spain.
Info graphic
Decision flow
Calibrated probabilities.
Info graphic
Friction map
Cost-based threshold.
Info graphic
Journey shape
Error analysis.
Next projects:
MLOPSFASTAPIDOCKERMONITORING
Credit Risk Platform
End-to-end MLOps platform for credit-default scoring: reproducible training with a versioned model registry, a FastAPI inference API with input validation, PSI drift monitoring and a Prometheus and Grafana observability stack, all containerised with CI. Code at github.com/delcenjo/credit-risk-platform.
View project
PYTORCHTRANSFORMERSNLP
Transformer from scratch
A GPT-style language model implemented from scratch in PyTorch: multi-head attention, causal masking and residual blocks written by hand, plus a byte-pair tokenizer and an ablation study. Code at github.com/delcenjo/transformer-from-scratch.
View project
NEXT.JSWEBASSEMBLYCLOUDFLARECI/CD
This very site
This portfolio is hand-built with Next.js and deployed on Cloudflare Workers, with a model lab that runs entirely in the browser and continuous deployment on every commit. Code at github.com/jmweb-org/jmweb-pag-web.
View project
SCIKIT-LEARNCALIBRATIONTABULAR
Credit Risk Model
Credit-default risk model with rigorous evaluation: a leakage-free pipeline, cross-validated model comparison, calibrated probabilities, a cost-based decision threshold and per-segment error analysis. Code at github.com/delcenjo/credit-risk.
View project