Deepak Sharma
Applied Scientist @ Microsoft • AI Agents • AI Evals • Post-training LLMs
Hi, I am an Applied Scientist at Microsoft, working on AI Agents to automate tasks in large businesses.
Current Focus
I own the AI Evals and quality strategy for our document-to-structured data extraction agent that powers Form Fill in PowerApps - leading synthetic data generation pipeline design, evals & metric design, error analysis, A/B experimentation, and curating data for and running SFT experiments. I am also exploring how I can make its tasks verifiable for reinforcement learning training for multi-turn, multi-tool use scenarios.
Research Background
I did ML research during my Master's at McGill University and MILA, where I worked under Dr. Joelle Pineau and Dr. Audrey Durand. I primarily focussed on model interpretability for ML applications in healthcare. I also worked on graph representation learning, model uncertainty, active learning, and causal inference, with applications aimed at improving human productivity and decision making in domains like healthcare and finance.
Contact
If you'd like to chat about research, discuss a potential collaboration or a job opportunity, please feel free to reach out to me via email or linkedin.