Ryan Russon

Capital One

• Owns the company's premier platform for orchestrating large-scale model-training jobs on Kubernetes, scaling to thousands of distributed nodes with Spark, Dask, and GPU PyTorch workloads across multiple clusters.
• Stood up guardrailed AI Sandbox and Prototyping environments that let internal customers build and experiment with the latest LLMs and agentic frameworks — provisioning and supporting dependencies such as LangGraph — turning frontier capabilities into production-ready, self-serve tooling.
• Leads MLOps in a highly regulated financial environment — vulnerability remediation, cluster access, and security — while preserving a smooth developer experience for model builders.
• Partners directly with scientists, engineers, and executives to translate ambiguous, cross-team requirements into shipped platform capabilities.

Maven Wave Partners (Google Cloud Consultancy)

• Embedded with a global bank to design and ship production model-training and -serving solutions on AWS using Kubeflow — including custom SDKs, CI/CD, and testing procedures for real-time fraud-detection models (XGBoost) on terabytes of data.
• Built an end-to-end training platform on Google Cloud for a delivery-forecasting system, orchestrating Vertex AI, BigQuery, and GCS via Kubeflow Pipelines with Stackdriver logging — cut model training time by 80% and improved prediction performance 4–10% over the client's model.
• Designed a fuzzy-matching system (Decision Tree + Connected Components) on AWS (Glue, Lambda, DynamoDB, RDS, SNS, SQS) resolving matches against billions of records and reducing execution time by 96%.
• Delivered a topic-modeling POC (Streamlit, AWS Transcribe, SageMaker, S3) surfacing major call-center themes to inform leadership on recurring, high-profile support issues.

Micron Technology / Intel Corp

• Developed deep-learning models (VGG-16, ResNet-152, transfer learning) for anomaly detection with automated responses, saving thousands of dollars weekly and preventing quality escapes.
• Built a custom human-in-the-loop web interface for rapid image labeling, enabling thousands of images to be classified in hours instead of days before model building.
• Created a Django web service solving constrained optimization for automatic process-control models, reducing human error by 98% and saving over $1M per year.
• Integrated relational (MSSQL/Oracle/Postgres/Hive) and NoSQL (Spark/HDFS/HBase) sources to deliver new manufacturing analytics; modernized ETL to improve data accessibility by over 300%.

U.S. Navy

• Managed the command's internal audit and continuous-improvement program spanning 60+ programs, 30 divisions, and 4,000 personnel; authored new training programs and policies.
• As a designated Navy Master Training Specialist (MTS), led an advanced course on the electrical and mechanical support systems for nuclear power-plant operation to 300+ personnel, achieving a 97% completion rate.

University of Utah

• Applied statistical methods (partial least squares, bootstrapping, ANOVA) to predict the onset of Alzheimer's disease from hundreds of MRI scans and cognitive assessments.

Agentic Guitar-Advice Web App (LangGraph, LLMs)

• Designed and built an agentic web application that delivers tailored guitar recommendations and advice, using LangGraph to orchestrate multi-step LLM reasoning over product and player-preference inputs.