Utkarsh Garg

About Utkarsh Garg

Principal Applied Scientist at Microsoft working on Word Copilot, where I build agentic LLM systems for long-form writing, precision editing, and reliable tool use.My core focus is making document AI controllable and measurable: I pioneered an HTML DOM–based editing framework for Word that enables targeted, incremental edits (instead of brittle regex or full rewrites), and I’ve built multi-agent platforms with state tracking, handoffs, persona-based user simulation, and synthetic data generation to run reproducible applied-science experiments. I also work on RAG for enterprise (intent/entity disambiguation, hybrid retrieval, reranking, multi-hop expansion, retrieval-budget control), RL fine-tuning + distillation for tool-calling reliability, and evaluation systems (including offline prompt A/B and LLM-as-judge validation).Before Microsoft, I led applied ML teams and shipped end-to-end systems at scale: • Built a customer-care chatbot (FAISS + Transformers + Siamese networks) and improved inference 85× using distillation, quantization, and ONNX. • Built mobile screen crack detection (CV pipeline) with monitoring + drift tracking, dramatically reducing onboarding time and operational load. • Developed RL-based recommendation approaches to improve engagement/retention, and predictive ML pipelines (XGBoost/ensembles) for ETA estimation.I enjoy turning research into shipped product, especially where reliability, evaluation rigor, and UX actually matter.

Experience

1. Principal Applied Scientist

   Microsoft

   Aug 2025 — Present · Seattle, WA, US

Education

• Delhi College of Engineering

  Bachelor of Technology (B.Tech.), Electrical and Electronics Engineering

  2010 — 2014

• University of California, Santa Cruz

  Master of Science - MS, NLP

Skills

• Business Intelligence
• Microsoft Sql Server
• Qlikview
• Data Warehousing
• Qlik Sense
• Requirements Analysis
• Business Requirements
• Data Modeling
• Python
• R
• Windows Server
• Data Analysis
• Embedded Systems
• Embedded C
• Solar Energy
• Teamwork
• C++
• Microsoft Excel
• Business Analysis
• Databases
• Hadoop
• Machine Learning
• Statistics
• Sql