Applied Scientist, Amazon Leo Satellite Build Systems
- Applied Science
- Applied Science
Train the language model that knows how Amazon builds satellites. As an Applied Scientist II, you'll own the science of the training corpus (selection, deduplication, filtering, enrichment, and data mixture) and of efficient inference (distillation and quantization) for Leo's domain-specific language model, so it answers manufacturing, quality, and test questions accurately at a serving cost the factory can sustain.
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
- Design the scientific components of the training-data pipeline: Source selection, deduplication within and across datasets, quality and perplexity filtering, poisoning detection, and sensitive-data screening
- Develop enrichment methods (summaries, diagrams rendered as text, structured representations, synthetic data, and rare-fact detection) and measure what each one adds to model quality
- Define data mixtures for continued pre-training and supervised fine-tuning, and run the ablations that justify them
- Build task-specific benchmarks and in-training evaluations that catch regressions before a model ships
- Work with scientists on reinforcement learning from verifiable rewards, and with engineers on training and serving infrastructure
A day in the life You start by reviewing last night's training run. An in-training evaluation shows a drop on test-procedure questions, and you trace it to a duplicated document cluster that crowded out rarer content. You write a cross-dataset deduplication pass and launch a small ablation. After lunch, you compare a distilled 4-bit model with the full-precision model on a nonconformance benchmark and find one failure class to fix before release. You end the day reviewing a teammate's enrichment experiment and updating the corpus bill of materials.
About the team Leo Intelligence Technologies is the central AI team in Leo Satellite Build Systems, and we're growing! We build shared AI capabilities for Production Operations: Governed data, a domain-specific language model, retrieval and knowledge services, evaluation frameworks, and agentic workflows that factory teams use every day. Our work changes real manufacturing decisions, so we value rigorous experiments, strong data foundations, clear documentation, and production-ready engineering. Scientists and engineers work side by side, and scientists take their work from experiment to production.
- Master's degree or above in computer science, machine learning, engineering, or related fields
- 3+ years of building models for business application experience
- Experience programming in Java, C++, Python or related language
- Experience in any of the following areas: algorithms and data structures, parsing, numerical optimization, data mining, parallel and distributed computing, high-performance computing
- Experience with Machine Learning and Large Language Model fundamentals, including architecture, training/inference lifecycles, and optimization of model execution, or experience in machine learning, data mining, information retrieval, statistics or natural language processing
openqareer не кадровое агентство: вакансия собрана с открытого источника, отклик подаётся на сайте работодателя.