Software Engineer II
Microsoft Search Assistant and Intelligence (MSAI)
Jun 2020 - Present
ML Infrastructure & Platform Engineering
• Architected and executed the end-to-end migration of an enterprise acronym service from on-box infrastructure to a multi-tenant Azure platform. Designed and deployed complex, multi-region ML inference workflows using Docker containers and Azure container orchestration. • Conducted systematic performance profiling across diverse Azure VM SKUs, optimizing resource allocation to meet strict request latency SLAs under high QPS loads. Instrumented the multi-region service with comprehensive telemetry to track per-CPU node metrics, throughput, and resource utilization patterns, leading to optimized operational efficiency. • Optimized time-based and event-based processors for email and meeting events by filtering for relevant user signals and
parallelizing data providers, reducing index rebuilding latency and operational cost. • Deployed production ML inference model replacing heuristic filters, improving recall by 15% while reducing data ingestion costs by 13%; implemented feature flighting for A/B testing model variants. • Designed scalable data processing pipelines handling ~25K daily user queries for M365 Copilot analytics, enabling failed query segmentation and user behavior analysis using LLMs.Search Optimization & Product Engineering
• Enhanced M365 Copilot meeting search relevance by implementing indexing support for attendance status, RSVP data, and external participants, expanding searchable metadata coverage. • Reduced search service P95 latency by 250ms through an advanced Warmup workflow to avoid JIT compilation overhead and caching frequently-accessed responses in machine plugin level memory cache. • Optimized backend query processing by parallelizing meetings data provider execution and batching and minimizing API payload sizes, reducing end-to-end response times. • Achieved $20K/annum in operational cost savings by optimizing distributed storage operations—replacing synchronous deletes with asynchronous expiration policies and batching transactions. • Migrated backend APIs from Service Fabric to Azure Functions, achieving a 50% performance increase, improved scalability, and 40% operational cost reduction. Built an Azure Data Factory (ADF) pipeline to ETL data from CosmosDb and SQL to
Azure Storage for consumption by analytics teams. • Integrated content sensitivity labels for Microsoft Information Protection (MIP) in Outlook Desktop, a feature commercialized under the M365 E5 license to enhance data security.
