Principal Applied Research Scientist – Generative AI and NLP

Yahoo Mail is the ultimate consumer inbox with hundreds of millions of users. It's the best way to access your email and stay organized from a computer, phone or tablet. With its beautiful design and lightning fast speed, Yahoo Mail makes reading, organizing, and sending emails easier than ever.

The Mail Intelligence team is the brain behind the inbox. We are responsible for building the next generation of platforms and services that enable Yahoo to deliver deeply personalized, intelligent, and context-aware experiences to hundreds of millions of users globally. We process billions of messages and manage data on a petabyte scale. Using cutting‑edge AI algorithms and foundation models , we extract knowledge and interconnect information from diverse sources to simplify our users' lives.

Building this knowledge provides many challenges in the areas of natural language processing, machine learning techniques , and petabyte-scale data processing. You will build tools and workflows to make it easier to manage and act on this vast information, applying your insights to build innovative consumer applications for Yahoo Mail.

You are a seasoned Applied ML Researcher who thrives at the intersection of theoretical innovation and production‑grade execution. You don't just follow the latest LLM trends ; you understand the mechanics of transformer architectures and how to optimize them for massive scale. You have expertise working across multiple ML and NLP spaces-including summarization, information extraction, classification, and ranking
-at a very large scale. You have hands‑on experience with knowledge distillation and believe that a model is only as good as its evaluation framework and low‑latency production performance. You combine strong research fundamentals with pragmatic production instincts, comfortably navigating ambiguity to drive high‑impact initiatives independently while mentoring and elevating engineering peers.

• Lead R&D:
  Drive the research strategy and development of deep learning and generative AI models specifically tailored for large‑scale email and communication data.
• Model Innovation & Optimization:
  Define and advance approaches for fine‑tuning and adapting open‑source foundation models using parameter‑efficient techniques (LoRA, adapters) and quantization‑aware training .
• Efficiency at Scale:
  Design and implement knowledge distillation to transfer complex capabilities into smaller, high‑performance models operating under strict latency budgets.
• Modern Evaluation Strategy:
  Establish and standardize robust evaluation frameworks, including LLM‑as‑a‑judge methodologies , synthetic evaluation datasets, and human‑in‑the‑loop validation.
• Product Integration:
  Build repeatable, scalable training and evaluation workflows for high‑throughput production environments.
• AI‑Augmented Development Workflow:
  Integrate AI pair‑programming tools (e.g., Copilot, Cursor) and automated LLM workflows to accelerate experimental iteration, code generation, and research prototyping.
• Strategic Future‑Proofing:
  Set technical direction for agent‑based systems, tool‑use paradigms, and long‑term generative AI roadmaps .
• Technical Mentorship & Governance:
  Raise the bar for technical excellence by guiding researchers, shaping org‑wide AI technical strategy, and establishing best practices for model validation.

• PhD (preferred) or Master's degree in Computer Science, Machine Learning, NLP, or a related field.
• 9+ years of hands‑on experience in applied machine learning and deep learning, with significant hands‑on work in NLP and generative models at scale.
• Demonstrated experience fine‑tuning LLMs using LoRA or other parameter‑efficient methods .
• Experience with knowledge distillation, model compression, and/or training smaller models from larger teacher models.
• Hands‑on experience using LLMs for pseudo‑labeling and synthetic data generation to create high‑quality training datasets.
• Experience building and evaluating LLM‑based agents, including tool use, multi‑agent orchestration, and reasoning workflows.
• Deep understanding of transformer architectures (encoder‑only,…