Maximiano Ezequiel

Led Generative AI and ML platform initiatives for enterprise data and AI infrastructure at global scale. Architected and deployed a Retrieval-Augmented Generation (RAG) reference architecture using LangChain and Databricks Vector Search, reducing customer proof-of-concept time by 45%. Migrated legacy model training pipelines to managed MLflow and serverless GPU compute with autoscaling, cutting customer compute costs by 30% while maintaining sub-second inference latency. Rebuilt the feature store ingestion layer (team of five), increasing throughput by 3x and enabling real-time personalization for Fortune 500 clients. Established remote-first mentorship that improved code review velocity by 25% and supported promotions within 18 months. Ensured enterprise security and compliance for generative AI reference implementations in regulated industries.

Focused on fraud detection and risk analytics for a cloud data platform serving thousands of enterprise customers. Built a real-time anomaly detection model with XGBoost and Snowpark, reducing false positives by 22% and preventing an estimated $4.5M in potential fraudulent transactions annually. Designed an experimentation framework to measure pricing elasticity, enabling product optimization of credit tiers and improving net revenue retention by 9%. Automated ETL workflows with Airflow and Snowpark to process 2TB of daily usage data, reducing data latency for billing dashboards from four hours to 15 minutes. Collaborated on a behavioral clustering model for coordinated account takeover detection with 91% precision. Mentored analysts on causal inference and created internal experiment design and statistical power analysis playbooks.

Developed computer vision and predictive modeling solutions for an IoT platform serving logistics and industrial customers. Fine-tuned a ResNet-based CNN for vehicle damage detection from fleet camera feeds, achieving 95% accuracy and reducing manual inspection time by 50%. Converted research models to ONNX and optimized inference on edge devices, reducing latency from 800ms to 120ms for real-time driver safety alerts. Partnered with U.S.-based product managers to define technical requirements and delivered three major features under aggressive deadlines while managing a five-hour time zone difference. Built a predictive maintenance model using gradient boosting to improve fleet uptime decisions and reduce average vehicle downtime by 14%. Documented and open-sourced an internal model deployment toolkit adopted by two other engineering squads.