Loading...
Senior Applied Scientist, Leo Satellite Build Intelligence
Amazon
On-siteseniorPosted May 13, 2026
About the Role
Design, train, and deploy ML models, including LLM-based systems, for production-critical decisions in satellite manufacturing, translate ambiguous problems into scientific solutions, and work with engineering teams for production integration.
Requirements
Lead the design, training, and deployment of advanced machine learning models, including LLM-based and generative AI systems, to create a closed-loop intelligence system for satellite manufacturing, addressing noisy and incomplete data.
Full Job Description
Build the scientific intelligence layer powering Amazon’s satellite manufacturing system. We are looking for a Senior Applied Scientist to lead the development of models that transform fragmented manufacturing, test, quality, and operational data into a unified, closed-loop intelligence system that directly improves how satellites are built.
You will work on high-ambiguity problems where data is incomplete, noisy, and distributed, and where model outputs directly influence real-world manufacturing decisions. Your work will power AI-native workflows such as non-conformance disposition, root-cause analysis, and predictive test optimization, reducing defects, accelerating production, and enabling self-improving manufacturing systems.
Export Control Requirement: Due to applicable export control laws and regulations, candidates must be a U.S. citizen or national, U.S. permanent resident (i.e., current Green Card holder), or lawfully admitted into the U.S. as a refugee or granted asylum.
Key job responsibilities
In this role, you will design and deploy purpose-built models that power production-critical decisions across satellite manufacturing.
- Lead the design, training, and deployment of machine learning models, including LLM-based systems, retrieval models, and task-specific models
- Translate ambiguous, real-world manufacturing problems into well-defined scientific problems, modeling approaches, and evaluation criteria
- Train, fine-tune, and evaluate models using large-scale, noisy, and heterogeneous datasets with incomplete or delayed ground truth
- Develop models over partially observed systems spanning test data, inspection signals, quality records, supplier data, and knowledge systems
- Invent and extend approaches for problems such as anomaly detection, root-cause inference, multimodal learning, and generative AI under real-world constraints
- Define evaluation frameworks that capture real-world failure modes, distribution shift, and decision risk, and use them to drive model iteration
- Make principled tradeoffs between model complexity, data quality, and generalization, and justify when to extend or depart from state-of-the-art approaches
- Work closely with engineering teams to deploy models into production systems with monitoring, feedback capture, and continuous retraining
- Build closed-loop learning systems where model outputs influence design, manufacturing, and test decisions
- Influence scientific direction across teams and mentor scientists and engineers
A day in the life
You may start by partnering with Quality, Manufacturing, and engineering teams to define and scope a training dataset for a root-cause prediction model, curating labels from historical cases. You then design and execute experiments to train and fine-tune models, comparing approaches across architectures, features, and data slices. Later, you analyze benchmark results, identifying failure modes, bias, and generalization gaps, and refine evaluation datasets to better reflect real-world edge cases. You iterate on model design and data quality before deploying the highest-performing model into a production workflow with monitoring, feedback capture, and retraining.
About the team
Leo Intelligence Technologies (LIT) is the centralized AI team within Leo Satellite Build Systems. We build the shared foundation for AI across Production Operations, including governed data assets, models, retrieval systems, evaluation frameworks, and knowledge services.
We operate on real-world systems where model outputs directly influence physical outcomes. We treat evaluation, data quality, and model behavior as first-class problems, and hold a high bar for rigor, auditability, and production readiness. Our work sits at the center of a shift toward AI-native manufacturing, where data, models, and feedback loops continuously improve production outcomes.
You will work on high-ambiguity problems where data is incomplete, noisy, and distributed, and where model outputs directly influence real-world manufacturing decisions. Your work will power AI-native workflows such as non-conformance disposition, root-cause analysis, and predictive test optimization, reducing defects, accelerating production, and enabling self-improving manufacturing systems.
Export Control Requirement: Due to applicable export control laws and regulations, candidates must be a U.S. citizen or national, U.S. permanent resident (i.e., current Green Card holder), or lawfully admitted into the U.S. as a refugee or granted asylum.
Key job responsibilities
In this role, you will design and deploy purpose-built models that power production-critical decisions across satellite manufacturing.
- Lead the design, training, and deployment of machine learning models, including LLM-based systems, retrieval models, and task-specific models
- Translate ambiguous, real-world manufacturing problems into well-defined scientific problems, modeling approaches, and evaluation criteria
- Train, fine-tune, and evaluate models using large-scale, noisy, and heterogeneous datasets with incomplete or delayed ground truth
- Develop models over partially observed systems spanning test data, inspection signals, quality records, supplier data, and knowledge systems
- Invent and extend approaches for problems such as anomaly detection, root-cause inference, multimodal learning, and generative AI under real-world constraints
- Define evaluation frameworks that capture real-world failure modes, distribution shift, and decision risk, and use them to drive model iteration
- Make principled tradeoffs between model complexity, data quality, and generalization, and justify when to extend or depart from state-of-the-art approaches
- Work closely with engineering teams to deploy models into production systems with monitoring, feedback capture, and continuous retraining
- Build closed-loop learning systems where model outputs influence design, manufacturing, and test decisions
- Influence scientific direction across teams and mentor scientists and engineers
A day in the life
You may start by partnering with Quality, Manufacturing, and engineering teams to define and scope a training dataset for a root-cause prediction model, curating labels from historical cases. You then design and execute experiments to train and fine-tune models, comparing approaches across architectures, features, and data slices. Later, you analyze benchmark results, identifying failure modes, bias, and generalization gaps, and refine evaluation datasets to better reflect real-world edge cases. You iterate on model design and data quality before deploying the highest-performing model into a production workflow with monitoring, feedback capture, and retraining.
About the team
Leo Intelligence Technologies (LIT) is the centralized AI team within Leo Satellite Build Systems. We build the shared foundation for AI across Production Operations, including governed data assets, models, retrieval systems, evaluation frameworks, and knowledge services.
We operate on real-world systems where model outputs directly influence physical outcomes. We treat evaluation, data quality, and model behavior as first-class problems, and hold a high bar for rigor, auditability, and production readiness. Our work sits at the center of a shift toward AI-native manufacturing, where data, models, and feedback loops continuously improve production outcomes.