Description
As a Principal Applied Scientist at Microsoft AI, you will lead the science behind Discover's ranking and content-quality stack, combining LLMs, multimodal models, and large-scale recommender systems to drive measurable gains in engagement, satisfaction, and trust. You will set technical direction, mentor a high-caliber science cohort, and partner closely with engineering, PM, UXR, and policy to ship end-to-end outcomes. You will contribute to the development of the next generation of MSN that is adopting the latest generative AI techniques.
Lead content-quality understanding at scale. Design and deploy models that assess credibility, usefulness, freshness, safety, and diversity across modalities; reduce misinformation/toxicity error rates through prompt- and model-level innovations; build human-in-the-loop and active-learning pipelines that get better over time.
Advance the recommendation & ranking stack. Architect and productionize large-scale DNN/LLM-enhanced recommenders (representation learning, sequence modeling, retrieval/ranking, slate optimization), balancing user satisfaction, content quality, and business goals.
Own evaluation and experimentation. Define offline metrics (e.g., NDCG, ERR, calibration) and online methodologies (A/B tests, interleaving, counterfactual & bandit approaches) to confidently attribute impact and guard against regressions.
Champion safety & trust. Partner with policy and platform teams to encode safety standards and editorial principles into the ML system; create red-teaming, adversarial, and safeguard layers for generative and curated experiences.
Scale E2E ML systems. Collaborate with engineering on data contracts, feature stores, distributed training/inference, and automated rollout/rollback; drive architectural investments that increase agility and reliability of Discover's AI platform.
Mentor & influence. Provide technical leadership across problem framing, methodology selection, code quality, and publishing/knowledge-sharing; uplevel peers through design reviews, deep-dives, and principled decision-making.
Stay close to users. Translate user engagements and behavioral history into model objectives and product bets; ensure our AI solutions elevate relevance, transparency, and engagement for real users.