Weights and Biases, Inc —
AMMO

The DoD requires a commercial MLOps infrastructure to sustain the operational utility of machine learning models in rapidly changing environments. This software pipeline must support post-deployment monitoring, efficient data exfiltration, rapid retraining, and hardware-specific model compilation. By integrating these CI/CD steps, the DoD ensures that models can be validated and redeployed quickly to account for adversarial activity and varying mission profiles.

Weights & Biases provided an MLOps pipeline that enables the rapid development, retraining, and deployment of machine learning models for autonomous explosive ordnance disposal tasks. The solution includes comprehensive software for data exploration, version control, model management, and experiment tracking, allowing for easy visualization of training datasets and metrics. This prototype successfully demonstrated the ability to rapidly retrain a target recognition model in a secure cloud environment to identify new threats in operationally relevant timelines.