Arlen Minas

Led cross-functional teams (10+ engineers) delivering production-grade AI systems across healthcare and finance, owning the full lifecycle from R&D to global scale. Engineered distributed training/inference infrastructure for 70B+ parameter models, optimizing GPU kernels and memory management to achieve peak GPU utilization. Designed multi-agent RAG systems using LangGraph and AutoGen, reducing manual processing time by 65%. Implemented LoRA/QLoRA-style fine-tuning to adapt foundation models for domain tasks, reducing cloud compute costs by 40%. Defined engineering standards for validation, monitoring, and security while mentoring senior staff and driving technical excellence.

Developed and deployed computer vision models using YOLOv5 for automated visual inspection, improving detection precision by 15%. Engineered a scalable RAG-based document retrieval system using ChromeDB and LangChain, achieving sub-second query latency for enterprise internal knowledge bases. Led the transition from monolithic ML scripts to a modular MLOps pipeline using Kubernetes and Docker, reducing deployment friction and improving environment consistency. Designed and managed distributed data processing jobs using Apache Spark and Kafka to enable real-time feature engineering for live inference services. Mentored junior engineers on robust model monitoring practices to detect data drift in production.

Built and deployed computer vision models using YOLOv5 for automated visual inspection, improving detection precision by 15%. Engineered a scalable RAG-based document retrieval system using ChromaDB and LangChain to enable sub-second query latency across enterprise internal knowledge bases. Led migration from monolithic ML scripts to a modular MLOps pipeline using Kubeflow and Docker, reducing deployment friction and improving environment consistency. Designed distributed data processing jobs with Apache Spark and Kafka to support real-time feature engineering for live inference services. Mentored engineers on robust model monitoring to detect data drift in production.