Staff Engineer, Hivemind Reasoning (R6292)

Shield AI•San Mateo, CA

About The Position

Shield AI is hiring a Staff Engineer for its Hivemind Reasoning team to build reasoning and cognition models for unmanned autonomous systems across aerial, maritime, land, and space domains. The engineer will drive model design, training, adaptation, and evaluation, translating emerging research into dependable capabilities for deployment on unmanned platforms. The work supports both individual vehicles and coordinated teams of hundreds of agents operating with limited compute, intermittent communications, incomplete information, and other edge constraints in contested environments. The role involves hands-on ownership of complex technical problems, from model architectures and training objectives to evidence of mission effectiveness. The engineer will collaborate with perception, MLOps, large-scale simulation, autonomy, hardware, and inference teams to deliver capabilities as part of Hivemind Enterprise. Responsibilities include comparing foundation-model approaches with specialized models and traditional AI/ML, building reproducible training workflows, evaluating models in simulation and with real-world data, diagnosing failures, and communicating results and limitations to technical and organizational stakeholders.

Requirements

Bachelor’s or master’s degree in Computer Science, Electrical Engineering, Mechanical Engineering, Aerospace Engineering, or a similar field, or equivalent practical experience.
Typically, at least 7 years of related experience with a bachelor’s degree, 6 years with a master’s degree, or 4 years with a PhD; equivalent work experience is also accepted.
Demonstrated track record of developing and deploying advanced ML models in production environments, with substantial hands-on contributions to model design, training, and evaluation.
Deep expertise in foundation models, modern deep learning, model architectures, and the training ecosystem.
Strong command of training and adaptation approaches such as self-supervised learning, supervised fine-tuning, imitation learning, reinforcement learning, and distillation.
Experience defining datasets, training objectives, experimental workflows, benchmarks, and reproducible criteria for comparing models.
Fluency in traditional AI/ML techniques and ability to combine them thoughtfully with learned models.
Experience with training frameworks such as PyTorch, JAX, or TensorFlow.
Experience with multimodal, embodied, robotic, or autonomous-system models; familiarity with VLA models and world models is particularly relevant.
Familiarity with traditional robotics algorithms, such as path planning, probabilistic modeling, and optimization approaches.
Familiarity with NVIDIA Jetson and other edge compute platforms, such as Qualcomm and AMD.

Nice To Haves

Master’s or PhD in Computer Science, Electrical Engineering, or a similar field.
Expertise in optimizing and deploying ML models to resource-constrained hardware platforms.
Experience with simulation-to-real on autonomous systems.
Experience with unmanned autonomous systems in defense applications.
Familiarity with large, distributed systems.
Familiarity with high-fidelity simulation, such as rigid body dynamics and sensor and actuator models.

Responsibilities

Design, implement, train, adapt, and evaluate reasoning and cognition models for unmanned autonomous platforms.
Make and validate tradeoffs among foundation-model approaches, including vision-language models (VLMs) and vision-language-action models (VLAs), specialized deep-learning models, and traditional AI/ML methods.
Develop capabilities for interpreting context, reasoning over time, coordinating among agents, and adapting to incomplete or changing information.
Build reproducible training and adaptation workflows using appropriate combinations of self-supervised learning, supervised fine-tuning, imitation learning, reinforcement learning, and distillation.
Define data requirements, training objectives, benchmarks, and evaluation methods that support meaningful comparisons among model approaches.
Establish rigorous evaluations in simulation and with real-world data, measuring generalization, robustness, uncertainty, and performance as team size and operating conditions change.
Incorporate edge constraints into model architecture and training decisions in partnership with hardware, autonomy, and inference teams.
Diagnose model failures and create feedback loops among data curation, training experiments, and re-evaluation.
Deliver trained models, reproducible results, and clear performance evidence to partner teams.
Communicate technical tradeoffs, experimental findings, and model limitations clearly to research, engineering, product, and organizational leadership.