Software Engineer - ML Infrastructure

Anduril's core Lattice platform integrates heterogeneous sensors across large fleets of assets to build a single common operating picture, providing mission-critical situational awareness and context to our users. As a Software Engineer on the ML Infrastructure team, you'll be responsible for developing the core underlying systems and infrastructure that perception engineers across the business leverage to field sophisticated, autonomous edge systems.

In particular, you'll develop our core data management systems for perception and autonomy use-cases, including indexing multi-modal sensor data, our model training/orchestration systems (that run across cloud-based and on-premise compute), as well as our internal perception platform. This platform enables dataset querying and discovery, orchestrating an internal suite of frontier models to tag and curate data, and ultimately train & evaluate models and deploy them back to edge devices.

This team also owns our core training libraries/SDKs for perception engineers, and partners with them to integrate novel model architectures into our standard infrastructure so the entire company can benefit from each improvement.

- Build & operate our core model training & dataset management infrastructure : Perception engineers rely on reliable, high quality data to continually retrain and improve our detection & tracing systems. You'll own and operate our core training systems, as well as our dataset management infrastructure.

- Improve perception engineer-facing tooling & SDKs: Perception engineers across Anduril rely on our core model training and evaluation framework. You'll partner closely with these engineers to understand their pain-points & feature requests, and iterate with them directly to deliver them.

- Integrate frontier models for data augmentation, curation, and synthetic data generation: While this team is not ultimately responsible for delivering perception systems for particular products, you will be responsible for integrating frontier models at all stages of our dataset curation/annotation pipeline, as well as enabling semantic search across our broad data holdings.

- Enable observability & traceability of the full ML lifecycle: You'll build and develop systems that enable real-time monitoring of training jobs, tools to store/curate model evaluation reports, and our model registry. This integrates with our Fleet Management systems to continuously deliver model improvements to our edge systems.

- Eagerly engage with our customers : We partner closely with business line engineering teams to understand their specific data requirements and iteratively deliver solutions that meet their particular needs and constraints.

- Strong proficiency in modern Python & exposure to Rust/the Rust ↔ Python toolchain (PyO3, Maturin, etc).

- Experience with modern deep learning frameworks such as Tensorflow or PyTorch, as well as distributed training paradigms

- Experience leveraging modern MLOps tooling on the orchestration side (Modal, Flyte, Airflow, Dagster, etc.), as well as common ML observability platforms (Weights & Biases, MLFlow, etc)

- Familiarity with building and deploying backend services into a Kubernetes environment

- U.S. Person status is required as this position needs to access export controlled data. Eligibility to obtain/maintain a US Top Secret clearance is also desirable.