Portrait of Tanuja Ganu

Tanuja Ganu

Director of Research Engineering

About

Hello! I am a Director of Research Engineering at Microsoft Research India, working at the intersection of multimodal AI, multilingual NLP, and agentic safety. My research focuses on building AI systems that perceive faithfully, reason reliably, and deploy responsibly across languages, modalities, and real-world contexts.

Previously, I led MSR’s center for Societal impact through Cloud and Artificial Intelligence (SCAI) that focused on creating, nurturing and deploying technologies that will have large scale impact on society. Prior to MSR, I co-founded DataGlen Technologies as CTO (an AI-powered monitoring and optimization platform for new-age energy systems with 800+ MW resources across 3 continents). Prior to this, I have worked as Research Engineer at IBM Research, India where I co-invented nPlug (opens in new tab) for decentralized energy demand management. I hold MS in Computer Science (Machine Learning) from Indian Institute of Science (IISc, Bangalore).

Research Areas

1. Multimodal AI
Understanding where and why vision-language models fail in perception, spatial reasoning, and cross-modal composition through diagnostic benchmarks (Do You See Me, Mind’s Eye, MMNeedle), and developing constrained optimization methods for grounded multimodal reasoning and complex visual reasoning over images and long-form videos.

2. Agentic AI Innovation
Building the next generation of AI agents that can plan, reason, and execute reliably over extended interactions, including self-evolving agents that improve through execution feedback, long-running agents with robust memory and context management, reliable multi-step reasoning with tool use, and safe multi-agent collaboration. This includes evaluation frameworks for adversarial robustness (TAMAS), safety of autonomous agents (SafeAgents), and prompt optimization (PromptWizard).

3. Population-Scale AI Deployments
Translating research into AI systems that serve real users at scale across languages, cultures, and resource-constrained settings: Shiksha Copilot for education (deployed for secondary and higher secondary Government school teachers across Telangana,), Rad-Phi4-Vision-CXR for medical radiology (#1 on Stanford’s RexRank VQA challenge), and Project Gecko for equitable AI across India and Africa with reusable deployment playbooks.