Applied Scientist III, AFT AI, Amazon AFT AI
Role details
Job location
Tech stack
Job description
In this role, you will build agentic AI solutions and multi-modal deep learning models that understand how products and packages flowing through Amazon's fulfillment network. You will build models that solve challenging problems like understanding warehouse operations systems, or visual defect detection on Amazon's entire retail catalog (billions of different items, thousands of new items every day). You will work with a diverse set of very large multi-modal real-world datasets, including imagery, natural language and structured data. You will face a high level of research ambiguity and problems that require creative, ambitious, and inventive solutions.
A day in the life AFT AI delivers the AI solutions that empower Amazon's fulfillment network to make smarter decisions. You will work on an interdisciplinary project involving scientists and engineers with deep expertise in developing state-of-the-art AI solutions at scale. You will work with images, videos, natural language, and sequences of events from existing or new hardware. You will adapt state-of-the-art agentic AI, deep learning, language understanding and computer vision techniques to develop solutions for business problems in the Amazon Fulfillment Network.
About the team Amazon Fulfillment Technologies (AFT) powers Amazon's global fulfillment network. We invent and deliver software, hardware, and science solutions that orchestrate processes, robots, machines, and people. We harmonize the physical and virtual world so Amazon customers can get what they want, when they want it.
Requirements
5+ years of relevant, broad research experience after a PhD degree or equivalent qualification
- Track record of first-author publications at top-tier peer-reviewed conferences (NeurIPS, ICML, ICLR, CVPR, ICCV, ECCV, ACL, EMNLP) or patents in machine learning domains
- Expert-level programming proficiency in Python with production-quality code standards, plus working knowledge of C++ for performance-critical applications; deep technical expertise with PyTorch and proficiency with the modern ML stack (Pandas, NumPy, scikit-learn, Hugging Face Transformers)
- Proven ability to independently scope, design, and execute end-to-end ML projects from research through production deployment, including ownership of model monitoring, maintenance, and iterative improvement
- Proven expertise in modern deep learning architecture design including transformers, diffusion models, and neural architecture search, with hands-on experience in designing and training self-supervised learning paradigms, training optimization techniques (distributed training across multi-node GPU clusters, mixed precision, gradient accumulation, parallelism strategies using DeepSpeed, FSDP, or Megatron-LM), and model compression methods (quantization, pruning, distillation)
- Proven experience pre-training and fine-tuning large language models (GPT, LLaMA, Claude) and vision-language models (CLIP, LLaVA, Qwen)
- Proven experience developing agentic AI systems deployed to production, using state-of-the-art frameworks (LangChain, Strands, etc.) with proven ability to design multi-agent workflows, tool-augmented reasoning systems, RAG systems and advanced prompt engineering techniques (chain-of-thought, few-shot, RLHF, DPO)
- Extensive knowledge and proven production experience across multiple ML domains including computer vision (object detection, segmentation, 3D vision, depth estimation, point cloud processing), natural language processing (text generation, information extraction), and multimodal learning
- Strong understanding of ML systems design including model serving infrastructure, A/B testing frameworks, feature stores, and MLOps best practices, such as annotation pipeline design, active learning pipelines, and AutoML/hyperparameter optimization techniques, Hands-on experience with cutting-edge generative AI techniques including diffusion models for image/video synthesis, autoregressive models for multimodal generation, and compositional generation systems; expertise in controllable generation, style transfer, and neural rendering techniques
- Deep expertise in model interpretability and explainability methods (attention visualization, feature attribution), with proven experience deploying interpretable AI systems in regulated or high-stakes production environments
- Experience with specialized ML domains such as few-shot learning, meta-learning, continual learning, or domain adaptation; proven ability to build models that generalize across distribution shifts, handle long-tail scenarios, or adapt to new tasks with minimal data
- Proven experience designing large language models (GPT, LLaMA, Claude) and vision-language models (CLIP, LLaVA, Qwen)
- Experience leading cross-functional ML initiatives involving multiple teams or organizations, with demonstrated impact on company-wide metrics or strategic product launches; proven track record of mentoring junior scientists and engineers in advanced ML techniques
- Published research contributions beyond first-authorship, including senior or corresponding author publications, invited talks at major conferences, or recognized leadership in ML research communities (program committee service, workshop organization, tutorial presentations)