微软Principal Applied Scientist

• Owns the science roadmap for grounding—including retrieval, re-ranking, attribution, and reasoning—driving initiatives from problem framing to production impact. Designs and evolves state-of-the-art retrieval and RAG orchestration across documents, tables, code, and images. • Builds citation and provenance systems (e.g., passage highlighting, quote-level alignment, confidence scoring) to reduce hallucinations and increase user trust. Leads experimentation and evaluation using A/B testing, interleaving, NDCG, MRR, precision/recall, and calibration curves to guide measurable trade-offs. • Advances tool-augmented grounding through schema-aware retrieval, function calling, knowledge graph joins, and real-time connectors to databases, cloud object stores, search indexes, and the web. Partners with platform engineering to productionize models with scalable inference, embedding services, feature stores, caching, and privacy-compliant multi-tenant systems. • Nurtures collaborative relationships with product and business leaders across Microsoft, influencing strategic decisions and driving business impact through technology. Authors white papers, contributes to internal tools and services, and may publish research to generate intellectual property. • Bridges the gap between researchers (e.g., Microsoft Research) and development teams, applying long-term research to solve immediate product needs. Leads high-stakes negotiations to ensure cutting-edge technologies are applied practically and effectively. • Identifies and solves significant business problems using novel, scalable, and data-driven solutions. Shapes the direction of Microsoft and the broader industry through pioneering product and tooling work. • Mentors applied scientists and data scientists, establishing best practices in experimentation, error analysis, and incident review. Collaborates cross-functionally with PMs, research, infrastructure, and security teams to align on milestones, SLAs, and safety protocols. • Communicates clearly through design documentation, progress updates, and presentations to executives and customers. Contributes to ethics and privacy policies, identifies bias in product development, and proposes mitigation strategies.

• Design and implement advanced LLM-based architectures and agentic systems for real-world product scenarios.• Translate research breakthroughs into production-ready algorithms, contributing to core capabilities such as reasoning, planning, long-term memory, code-gen based design.• Monitor and improve model performance post-deployment through data-driven iteration and error analysis.• Collaborate across teams to deliver robust, scalable models aligned with product objectives and user value.• Contribute to the organization’s scientific direction by identifying research opportunities that drive long-term differentiation.

• Research, design, and prototype methods to leverage LLMs for product scenarios such as text understanding, summarization, dialogue, translation, content generation, and reasoning. • Fine-tune, adapt, and optimize pre-trained LLMs for domain-specific tasks while balancing model performance, efficiency, and cost. • Develop scalable pipelines for data collection, cleaning, augmentation, and evaluation. • Collaborate with product and engineering teams to translate applied research into production-quality features. • Define and track key performance metrics for LLM-based features, including accuracy, latency, robustness, and user satisfaction. • Stay current with advances in generative AI, multimodal models, and applied ML techniques, and bring forward innovative ideas to improve our products. • Publish technical insights internally (and externally where appropriate) to advance organizational knowledge and thought leadership.