Cody Freese

I’m an AI/ML Engineer focused on spatiotemporal detection and behavior classification at GEOINT scale, turning complex sensor data into reliable, decision-ready signals.

I built production AWS pipelines and feature pipelines for orbital elements, kinematics, and anomaly-detection prototypes that surface non-nominal maneuvers for GEOINT triage, using rigorous metrics like ROC-AUC, log-loss, PR/AUPRC, and calibration.

On Transformer-based multi-object tracking, I implemented DETR-style object/track queries with joint detection+association, evaluated with HOTA/IDF1/MOTA, and supported masked training, early stopping, and gradient clipping—plus modular loaders for variable-length sequences.

I co-authored BATMAN and developed EO–AIS fusion pipelines, including behavior labeling, SHAP explainability, and simulation-driven stress testing for sensor latency, revisit rate, message loss, and data-fusion timing to enable reproducible regression and calibration for downstream ML workflows.