AI Robustness and Reliability

Building dependable AI for applications that benefit people and society.

My research spans conversational AI, agentic systems, language-model reasoning, and responsible AI. Through work at Microsoft Research, DoorDash, and UdeM, I’ve been drawn to building AI that addresses practical problems and has meaningful societal impact.

The variability of language and multimodal models raises questions about how dependably they can support real-world tasks. These questions have shaped my growing interest in AI robustness and safety, and motivate my current research on making model behavior more reliable.

Research interests

• Conversational and agentic AI: dialogue, reasoning, and agents that help people with real-world tasks.
• AI robustness and safety: understanding failure modes and improving the reliability of language and multimodal models.
• AI for social impact: applications that benefit people, with attention to fairness and the communities they affect.

For concrete examples, explore COSMIR’s structured memory for long-context reasoning and fairness in preference learning. For applied research, see my DoorDash and Microsoft Research experience.

Working on a related question? Let’s talk →