Staff Gen AI Research Scientist

ABOUT THE TEAM

The Air Dominance & Strike team at Anduril develops aerial and multi-domain robotic systems. The teamis responsible fortaking products like Fury (unmanned fighter jet) and Barracuda (air-breathing cruise missile) from concept to product. The team also develops Lattice for Mission Autonomy Andurils premier software platform that enables masses of Fury Barracuda and other first andthird partyrobots to collaborate across various missions. We work in close coordination with specialist teams like Perception Motion Planning Hardware and Test Engineering to solve some of the hardest problems facing our customers. We are looking forAI research scientistsAppliedscientistsand roboticists excited about creating a powerful autonomy software stack that includes computer vision motion planning SLAM controls estimation and secure communications.

ABOUT THE JOB

We are seeking anAI Research Scientistto serve as a founding ML expert on our this role you will design fine-tune and deploy the next generation of generative AI LLMs and agentic systems that power our air-dominance platforms and collaborative autonomous behaviors.

This isa highly appliedresearch role(splitroughly60%applied research/experimentation and40% hands-on coding)focused on makingstate-of-the-artLLMmodels smaller faster and smarter. You will work on both offboard systems (for complex mission planning modeling and simulation) and onboard systemsoptimizingmodels to run directly on power- and compute-constrained edge hardware. As an early member of this initiative you will have significant autonomy to set the technical direction design our data collection strategy across test sites and simulations and directly influence how multi-agent autonomy is deployed in critical missions.

WHAT YOULL DO

• Develop pre-train and fine-tune in-house LLMs and multimodal foundation models. Apply SOTA post-training alignment techniques (SFT RLHF DPO) to maximize capability while minimizing cost and footprint.
• Architect and optimize models to run directly on tactical edge compute and power-constrained hardware onboard physical assets. Optimize model latency memory usage and execution speed through quantization distillation and pruning.
• Design and implement robust agentic architectures multi-agent coordination frameworks and planning loops for complex multi-domain military missions.
• Collaborate closely with computer vision perception and motion planning teams to build systems capable of reasoning over diverse modalities including camera feeds radar telemetry and text-based operational orders.
• Define and execute data collection strategies across physical assets test sites and virtual simulations. Work with AI Infrastructure engineers to build scalable evaluation frameworks that measure model performance reliability and safety in high-stakes environments.
• Build early-stage prototypes alongside customers quickly iterate on feedback and scale those prototypes into production-grade features deployed across our family of systems.

REQUIRED QUALIFICATIONS

• Strong production-level coding skills in Python and deep learning frameworks (likePyTorchor JAX).
• Hands-on experience training fine-tuning and evaluating LLMs Generative AI or multimodal models.
• A strong background in a classical technical discipline (Computer Vision NLP Robotics or Speech) with 2 years of dedicated experience focusing on generative models and modern transformer architectures.
• Experience using modern model training alignment and orchestration tools (e.g. Axolotl Hugging Face DeepSpeed Megatron-LM LangChain or LlamaIndex).
• Ability to operate comfortably in a fast-paced environment moving from ambiguous mission requirements to concrete code and functional prototypes.
• Degree (B.S. M.S. or Ph.D.) in Computer Science Machine Learning Robotics Physics Mathematics or a related technical field.
• Eligible to obtain and maintain an active U.S. Top Secret security clearance.

PREFERRED QUALIFICATIONS

• Proventrack recordof compiling and running deep learning models on edge accelerators (e.g. NVIDIA Jetson custom TPUs/ASICs) under severe power and compute constraints.
• Hands-on experience implementing Reinforcement Learning from Human/AI Feedback (RLHF/RLAIF) or direct preference optimization (DPO) loops.
• Experience working with multimodal architectures (VLM video-to-text or sensor fusion) and diffusion models.
• Experience building and shipping AI-powered products used by millions of users with a deep understanding of the end-to-end lifecycle from research prototype to production deployment.
• Track record of successfully applying AI/ML technologies to novel problems and domains where established approaches do not yet exist demonstrating creativity and first-principles thinking.
• Proven ability to rapidly learn new technologies and stay current with the fast-evolving AI landscaperegularly engaging with the latest research tools and techniques through publications conferences and the open-source community.

The salary range for this role is an estimate based on a wide range of compensation factors inclusive of base salary only. Actual salary offer may vary based on (but not limited to) work experience education and/or training critical skills and/or business considerations. Highly competitive equity grants are included in the majority of full time offers; and are considered part of Andurils total compensation package. Additionally Anduril offers top-tier benefits for full-time employees including:

At Anduril we invest in our people. Our comprehensive competitive benefits package (available at little to no cost to employees) ensures youre supported in health recovery and whatever comes next.For more information Explore Our Benefits.

To view Andurils candidate data privacy policy please visit submitting your application you consent to Anduril Industries using a third-party service provider to conduct pre-employment risk integrity and due diligence screening and assessing potential risks as part of your application process. This third-party service provider provides risk-intelligence services that may include analysis of sanctions and watchlists adverse media public-record information and other lawful open-source or commercial data sources. This third-party service provider does not act as a consumer reporting agency. Use of this provider helps to ensure compliance with applicable laws and protect technology intellectual property and organizational security.