Principal Engineer, AI Infrastructure; R4941

Founded in 2015, Shield AI is a venture‑backed deep‑tech company with the mission of protecting service members and civilians with intelligent systems. Its products include the V‑BAT and X‑BAT aircraft, Hivemind Enterprise, and the Hivemind Vision product lines. With offices and facilities across the U.S., Europe, the Middle East, and the Asia‑Pacific, Shield AI’s technology actively supports operations worldwide. For more information, visit (Use the "Apply for this Job" box below).

Follow Shield AI on Linked In, X, Instagram, and You Tube.

Shield AI builds autonomy systems for defense applications, including air, maritime, and space platforms operating in complex and contested environments.

We are establishing a centralized AI and Data Platform organization responsible for the infrastructure that underpins autonomy development across Hivemind and other programs. This team owns the systems used to train models, run simulation, manage data, and deploy models to operational environments.

We are seeking a Principal Engineer that will scale an initial architecture into a platform that supports multiple autonomy programs.

Success in this role requires disciplined execution, delivering fast iteration for engineering teams while maintaining reliability, cost control, and architectural consistency as the system scales.

The Principal Engineer is accountable for ensuring engineers can move efficiently from idea to trained model to deployed capability, and that infrastructure decisions reflect the realities of the domain, including simulation‑driven development, continuously evolving multi‑modal sensor data, and deployment to constrained and reliability‑critical systems.

This role spans the full lifecycle of autonomy development, training foundation models, running large‑scale and multi‑fidelity simulation, managing training data, evaluating models, and deploying optimized models to edge systems.

A key part of this role is defining how these capabilities extend beyond internal use. This includes establishing how Shield AI delivers AI infrastructure in customer environments across on‑premise, cloud, hybrid, and sovereign or nationally constrained environments.

• Platform Ownership: Define and operate the core AI and data platform across training, simulation, data management, evaluation, and deployment.
• Compute Strategy and Infrastructure: Own where and how workloads run across on‑premise, cloud, and hybrid environments. Drive capacity planning, utilization, and cost‑per‑compute decisions, including support for classified and air‑gapped systems.
• Training and Simulation Systems: Build infrastructure for distributed training (supervised learning, RL/MARL, foundation models) and large‑scale, multi‑fidelity simulation. Ensure training and simulation systems operate together without bottlenecks.
• Data Platform: Ingest and manage multi‑modal sensor data (EO, IR, radar, EW, IMU). Establish dataset versioning, data lineage, feature storage, data cataloging, and classification‑aware storage and access controls.
• MLOps, Evaluation, and Model Lifecycle: Establish a consistent workflow for experiment tracking, model registry, artifact provenance, and automated validation. Implement evaluation and V&V gates so models meet defined standards before deployment.
• Deployment and Operational Feedback: Own the pipeline from training to deployment, including model optimization (e.g., distillation, quantization, pruning), deployment to edge systems, monitoring, drift detection, and retraining triggers.
• Customer AI Infrastructure: Define how AI infrastructure is deployed in customer environments across on‑premise, cloud, hybrid, and sovereign settings. Establish a consistent approach that avoids one‑off solutions while adapting to operational constraints.
• Platform Standardization: Define common tools, interfaces, and workflows across teams. Reduce duplication while maintaining flexibility where needed.
• Cross‑Team Partnership: Work directly with Hivemind and other autonomy teams to ensure the platform supports real workloads and evolves with program needs.

• Faster iteration from idea to trained model to evaluated result
• High…