Data Scientist & Machine Learning Engineer
El Paso, Texas, United States
I build machine learning systems that stay cheap enough to deploy and honest enough to trust. My research compresses large language models until they run on ordinary hardware, and my engineering work focuses on catching models when they quietly stop working.
A compact transformer for contextual log anomaly detection reaching 98.83% accuracy at roughly 70% lower compute cost than large-LLM baselines. First-author paper under review at IEEE Transactions on Neural Networks and Learning Systems.
A live champion/challenger console for card-fraud detection: Population Stability Index drift monitoring, adaptive traffic allocation, and model promotion gated by McNemar's test. Shows a model whose recall collapses from 97% to 25% — and the machinery that catches it.
How many labels must you buy before a supervised intrusion detector beats an unsupervised one that needs none? Isolation Forest, gradient boosting and an RBF-SVM written from first principles in ~500 lines, no libraries, with a 36-check test suite.
Ph.D. in Data Science, University of Texas at El Paso. M.S. in Mathematics of Finance and Economics, University of Silesia. M.S. in Engineering Mathematics, University of L'Aquila. B.S. in Financial Mathematics, University for Development Studies. Previously a technical specialist in quantitative risk and data analytics at AT&T.