|

Building production ML systems with a focus on interpretability and measurable impact.

PythonKedroscikit-learnMLflow

FEATURED PROJECTS

Selected work

BizSentinel

Completed

End-to-end ML platform for e-commerce customer intelligence

97,896Customers processed
1.47%Anomaly rate
82%Test coverage
127Unit tests

Spot high-risk customers, anomalous transactions and churn before revenue leaks — 97,896 SMB e-commerce customers scored end-to-end.

PythonLightGBMScikit-learnKedroMLflowFastAPIDashDocker

Hardware Pulse

In Progress

PC hardware price intelligence for the Uruguayan market

3+Retailers integrated
4Component categories
3Resolution pipeline stages
PydanticData contracts

Helps Uruguayan PC buyers time their purchase and find the cheapest retailer across GPUs, CPUs, SSDs and RAM.

PythonBeautifulSouppandasSQLiteRapidFuzzScikit-learnStreamlit

Music Taste Recommender

Completed

Business-driven hybrid recommendation engine

+267%Diversity improvement
0.89Relevance maintained
217KTracks processed
500Simulated users evaluated

Helps streaming listeners discover more of the catalog without losing relevance — pushing genre diversity from 19% to 70% at a 0.89 relevance score, so users explore instead of churning out.

PythonSentence-TransformersScikit-learnStreamlitHugging Facepandas

TECHNICAL SKILLS

What I work with

Machine Learning

PythonScikit-learnLightGBMKerasSHAP

MLOps & Pipelines

KedroMLflowPrefectDockerFastAPI

Data Engineering

pandasSQLitePostgreSQLBeautifulSoupLinux

Frontend & Viz

StreamlitDash

EXPERIMENTS & NOTEBOOKS

Lab

Sports Image Classification

CNN-based image classifier trained to recognize sports categories from photos.

CNNComputer VisionKeras

Heart Disease Classification

SVM vs Random Forest comparison on 918 examples. SVM achieved F1-Score of 0.897 on test set.

SVMRandom ForestFeature SelectionClassification

Credit Card Fraud Detection

Fraud detection using clustering and anomaly detection on imbalanced transaction data.

Anomaly DetectionClusteringImbalanced Data