At Ciudades, I run data-quality improvement as a measurement program. I classified findings by root cause and built deterministic regression checks, cutting output noise from 43% to 20%.
I designed a pre-registered evaluation program and hand-validated its LLM-as-judge pipeline before using results. A prompt rewrite lifted defect recall from 55% to 75% at equal cost, while an agent-routing rule cut model spend by about 65%.
At the National Hydrocarbons Commission (CNH), I architected and managed a centralized PostgreSQL database that became the single source of truth for the Commissioner's advisory office. I also built ETL processes to reconcile production, reserves, CAPEX, and OPEX data before it reached executive deliverables.
I developed a Predictive Energy Model, moving it from an Excel/VBA prototype to a reproducible Python pipeline. My work has also included an investment analytics platform using Airflow, BigQuery, and dbt, and a customer segmentation and recommendation project using K-Means and PCA.
