I am an AI researcher and research engineer working on reliable AI systems. My work spans calibrated uncertainty for robot perception, structured procedural understanding, multimodal grounding, and, currently, agentic AI harnesses, evaluation, and post-training at a stealth startup. I am based in the United States and am a US permanent resident.
Previously, I worked at Samsung AI Centre Toronto on multimodal learning, image retrieval, graph parsing, LLM/agent prototypes, and on-device AI. Before Samsung, I was a research masters student at MILA, University of Montreal, under supervision of Professor Liam Paull, where I mainly worked on probabilistic object detection and trustworthy uncertainty estimation. Before coming to MILA, I did MS in Robotics with Prof. Madhav Krishna, and worked on self-driving cars, autonomous drones, and robot perception.
I like the parts of AI work where research taste meets engineering discipline: building harnesses, testing failure modes, turning fuzzy ideas into runnable loops, and making systems easier to measure.
Current focus
A short snapshot of the ideas I keep circling back to.
Agentic AI systems
Workflows that use tools, recover from errors, and expose measurable behavior.
Harnesses and evaluation
Controlled loops that make model progress visible instead of anecdotal.
Post-training
Using feedback, demonstrations, and synthetic data to shape more useful model behavior.
Fun Fact
The profile picture you see on this website is not a real photograph. I fine-tuned Flux1-dev on 17 images of myself and generated this image. Curious about the process? Check out the details here.