Staff Machine Learning Scientist, Agentic AI

PayPal•San Jose, CA

About The Position

Lead the development and optimization of advanced machine learning models. Oversee the preprocessing and analysis of large datasets. Deploy and maintain ML solutions in production environments. Collaborate with cross-functional teams to integrate ML models into products and services. Monitor and evaluate the performance of deployed models, making necessary adjustments. Design, develop, and evolve Commerce Agentic System, advancing its reasoning, memory, and action layers to create dynamic, context-aware user experiences. Fine-tune large language models (LLMs) for commerce and shopping applications, ensuring robust alignment, safety, and personalization. Implement and extend agentic frameworks such as A2A (Agent-to-Agent), MCP (Model Context Protocol), LangGraph, or similar to enable complex multi-agent interactions. Perform advanced context and prompt engineering, optimizing multi-turn, multi-source model orchestration for superior performance and responsiveness. Collaborate cross-functionally with product, design, and platform engineering teams to define the next generation of agentic capabilities and AI interface strategies. Experiment with reinforcement learning, retrieval-augmented generation (RAG), and online adaptation to refine agent behavior and enhance response quality. Build scalable, production-ready pipelines for model training, evaluation, and continuous improvement.

Requirements

5+ years relevant experience and a Bachelor's degree OR Any equivalent combination of education and experience.
Extensive experience with ML frameworks like TensorFlow, PyTorch, or scikit-learn.
Expertise in cloud platforms (AWS, Azure, GCP) and tools for data processing and model deployment.
Master's degree (or higher) in Computer Science, Artificial Intelligence, Machine Learning, or a related quantitative discipline.
5+ years of relevant industry experience (or 4+ years with a PhD).
Deep understanding of Transformer architectures and hands-on experience with fine-tuning LLMs for production use cases.
Strong proficiency in Python and ML frameworks such as PyTorch, TensorFlow, or JAX.
Demonstrated experience with Agentic frameworks such as A2A, MCP, LangGraph, LangChain or similiar, and an understanding of Agent-Oriented Design patterns.
Experience building context-aware conversational systems, integrating multi-source data for reasoning and response generation.
Knowledge of LLM and agentic evaluation methodologies, including prompt testing, offline metrics, and human feedback loops.
Familiarity with MLOps/LLMOps practices — model deployment, monitoring, and continuous retraining at scale.
Excellent communication skills with the ability to collaborate across engineering, research, and product teams.

Responsibilities

Lead the development and optimization of advanced machine learning models.
Oversee the preprocessing and analysis of large datasets.
Deploy and maintain ML solutions in production environments.
Collaborate with cross-functional teams to integrate ML models into products and services.
Monitor and evaluate the performance of deployed models, making necessary adjustments.
Design, develop, and evolve Commerce Agentic System, advancing its reasoning, memory, and action layers to create dynamic, context-aware user experiences.
Fine-tune large language models (LLMs) for commerce and shopping applications, ensuring robust alignment, safety, and personalization.
Implement and extend agentic frameworks such as A2A (Agent-to-Agent), MCP (Model Context Protocol), LangGraph, or similar to enable complex multi-agent interactions.
Perform advanced context and prompt engineering, optimizing multi-turn, multi-source model orchestration for superior performance and responsiveness.
Collaborate cross-functionally with product, design, and platform engineering teams to define the next generation of agentic capabilities and AI interface strategies.
Experiment with reinforcement learning, retrieval-augmented generation (RAG), and online adaptation to refine agent behavior and enhance response quality.
Build scalable, production-ready pipelines for model training, evaluation, and continuous improvement.