🧪 AI Labs

Artificial Intelligence, Machine Learning, Deep Learning and Computer Vision projects applied to real operational problems. Each project follows CRISP-ML methodology and is deployed in production.

Proyectos de Inteligencia Artificial, Machine Learning, Deep Learning y Computer Vision aplicados a problemas operativos reales. Cada proyecto sigue la metodología CRISP-ML y está desplegado en producción.

📐 Methodology: 📐 Metodología: CRISP-ML(Q) | 🚀 Deploy: Streamlit Cloud

🪤 ● Live

Time Series · Forecasting · MLOps

Availability Leakage — the CV winner you shouldn't ship

Leakage de Disponibilidad — el ganador del CV que no deberías shipear

CRISP-ML(Q) study on the Favorita dataset (Kaggle): transactions correlates 0.84 with sales and tops feature importance — but it doesn't exist at forecast time. The leak doesn't inflate a score, it inverts model selection. RMSLE gap +0.060 (47%).

Estudio CRISP-ML(Q) sobre el dataset Favorita (Kaggle): transactions correlaciona 0.84 con ventas y lidera la importancia de features — pero no existe al forecastear. El leak no infla un score, invierte la selección de modelo. Gap +0.060 (47%).

Python LightGBM Data Leakage CRISP-ML

🏪 Retail · Forecasting Retail · Forecasting

View Details → Ver Detalles → GitHub ↗

📦 ● Live

Demand · Censored Data · ML

Censored Demand — recovering the sales your ERP never sees

Demanda Censurada — recuperar la venta que el sistema no ve

CRISP-ML(Q) on FreshRetailNet-50K: 44% of days are censored by stockouts. Recovering latent demand cuts bias −0.132→−0.014 and reduces MAE 5.9%. Recovery is not a free lunch — tested across 10 model families.

CRISP-ML(Q) sobre FreshRetailNet-50K: 44% de los días censurados por quiebres. Recuperar la demanda latente baja bias de −0.132→−0.014 y reduce el MAE 5.9%. No es un free lunch — probado en 10 familias de modelos.

Python LightGBM statsforecast CRISP-ML

🛒 Retail · Supply Chain Retail · Supply Chain

View Details → Ver Detalles → GitHub ↗

💳 ● Live

Classification · Imbalanced · Responsible AI

South German Credit — Cost-Aware Risk & a Fairness Audit

South German Credit — Riesgo Sensible al Costo y Auditoría de Equidad

CRISP-ML(Q) on South German Credit (UCI). Threshold tuning alone (5:1 cost ratio) lifted recall 29%→84% and cut business cost 53% — same model, no retraining. Fairness audit: equalized-odds gap 0.62, disparate impact 0.41.

CRISP-ML(Q) sobre South German Credit (UCI). Solo el threshold tuning (ratio 5:1) subió el recall 29%→84% y bajó el costo un 53% — mismo modelo, sin reentrenar. Auditoría de equidad: equalized-odds gap 0.62, disparate impact 0.41.

Python XGBoost SMOTETomek fairlearn

🏦 Banca · Credit Risk Banca · Credit Risk

View Details → Ver Detalles → GitHub ↗

📈 ● Live

Time Series · Regression · ML

Retail Demand Forecasting — Two MAE Plateaus

Pronóstico de Demanda Retail — Dos Mesetas de MAE

CRISP-ML(Q) study on a Kaggle retail dataset: how the existing forecast enters the pipeline — residual prior vs dropped leakage trap — splits the holdout into two MAE plateaus (~69 vs ~7.4) on the same data and models.

Estudio CRISP-ML(Q) sobre un dataset retail de Kaggle: cómo entra el forecast existente al pipeline — prior residual vs descartado por leakage — parte el holdout en dos mesetas de MAE (~69 vs ~7.4) sobre los mismos datos y modelos.

Python LightGBM HGB Residual Learning Streamlit

🚛 Retail · Supply Chain Retail · Supply Chain

View Details → Ver Detalles → GitHub ↗

🔧 ● Live

Classification · ML

Equipment Failure Prediction

Predicción de Fallas en Equipos

Predictive maintenance system that classifies failure type before it occurs, optimizing maintenance costs and preventing unplanned downtime.

Sistema de mantenimiento predictivo que clasifica el tipo de falla antes de que ocurra, optimizando costos de mantenimiento y evitando paros no programados.

Python Scikit-learn XGBoost Streamlit

🏭 Manufacturing · Fleets Manufactura · Flotas

View Details → Ver Detalles → GitHub ↗

👥 ● Live

Classification · Binary · ML

HR Employee Attrition Prediction

Predicción de Rotación de Empleados

Machine learning system to identify employees at risk of leaving, enabling proactive retention strategies and reducing turnover costs.

Sistema de machine learning para identificar empleados en riesgo de irse, permitiendo estrategias de retención proactivas y reduciendo costos de rotación.

Python LightGBM XGBoost Streamlit

👔 HR · Talent Management RRHH · Gestión del Talento

View Details → Ver Detalles → GitHub ↗

👁️ Planned Planificado

Computer Vision · Deep Learning

Operations Object Detection

Detección de Objetos en Operaciones

Computer vision model for automatic detection and counting of objects in industrial environments: pallets, PPE, products.

Modelo de visión por computadora para detección y conteo automático de objetos en entornos industriales: pallets, EPP, productos.

Python YOLO OpenCV Streamlit

📦 Logistics · Security Logística · Seguridad Coming soon Próximamente

📐 CRISP-ML(Q) MethodologyMetodología

All projects follow the CRISP-ML(Q) standard for Machine Learning, ensuring reproducibility, documentation and quality at each phase.

Todos los proyectos siguen el estándar CRISP-ML(Q) para Machine Learning, garantizando reproducibilidad, documentación y calidad en cada fase.

🎯

Business Understanding

📊

Data Understanding

🔧

Data Preparation

🧠

Modeling

✅

Evaluation

🚀

Deployment