Computer Science · University of Toronto
I work on what happens to language models after pretraining: how reinforcement learning and distillation shape their reasoning, and how they behave as agents. This fall I start my MSc with Ashton Anderson in the Computational Social Science Lab.
I’m an incoming MSc student in computer science at the University of Toronto, where I work with Ashton Anderson on post-training for reasoning: reinforcement learning, on-policy distillation, and models that catch their own mistakes. I’m also interested in LLM agents and in language models for recommendation and search.
Before this I spent two years on recommendation and retrieval in Scott Sanner’s lab, studied LLM reasoning in Alán Aspuru-Guzik’s Matter Lab, and did stints at Air Canada and Bosch. I also co-founded ZBot Technology, where we build AI agents for exporters.
How far can reinforcement learning and on-policy distillation push reasoning? ThinkTwice trains models to draft an answer and then repair it. Chess is a small test bench for the same question: every claim a model makes there can be checked exactly.
An agent that reads the web and acts on your behalf can be steered by whoever wrote what it reads. SafeGEO measures how generative engine optimization distorts what recommendation agents suggest. Building agents at ZBot Technology keeps this research honest.
Language models change how people search for things they cannot name yet. EQR expands a vague query into the subtopics it implies, GPR-LLM scores items with Gaussian processes over LLM relevance judgments, BAGEL decides where each of those judgments is best spent, and MA-DPR rethinks what “close” means in embedding space.
Starting my MSc in Computer Science at the University of Toronto, supervised by Ashton Anderson.
SafeGEO is on arXiv: understanding generative engine optimization risks in recommendation agents, with a project page.
Completed my BSc in Computer Science at the University of Toronto (specialist, focus on AI).
Awarded the 2026–27 Vector Scholarship in AI by the Vector Institute.
Two of our papers were accepted to Findings of ACL 2026: Gaussian-process item scoring with LLM relevance judgments and Bayesian active learning for dense passage retrieval.
MA-DPR, our manifold-aware distance metric for dense passage retrieval, was accepted to the EMNLP 2025 main conference.
Received the Research Award of the Department of Computer Science, University of Toronto.
* denotes equal contribution
University of TorontoUniversity of Toronto
Computational Social Science Lab, University of TorontoPost-training for reasoning. I co-developed ThinkTwice, which jointly trains models to reason and to refine their own answers. We also use chess as a verifiable test bench: we distill master play into grounded reasoning and measure understanding with ChessQA.
Data-Driven Decision Making Lab, University of TorontoTwo years on natural-language recommendation and retrieval. I worked on elaborative query reformulation for queries that describe a need without naming an item, on item scoring with Gaussian processes over LLM relevance judgments (Findings of ACL 2026), and on manifold-aware distance metrics for dense passage retrieval (EMNLP 2025).
The Matter Lab, University of TorontoAsked whether language models actually reason. We probed multi-agent debate and tree-of-thought prompting, and tried compositional training for diffusion models on ARC-style puzzles.
We build AI agents for export businesses. I run the technical side.
Air CanadaBuilt a travel recommender for people who know the trip they want but not the place, and released TravelDest, the first benchmark dataset for travel search. The approach became a RecSys 2024 workshop paper.
BoschMotion planning for the Wave3 autonomous driving program. The Monte Carlo tree search framework I worked on there later became an IEEE SSRR 2024 paper.
Vector Scholarship in Artificial Intelligence · Vector Institute
Research Award · Department of Computer Science, University of Toronto
Reviewer · IEEE International Conference on Robotics and Automation (ICRA)
Chancellor’s Scholarship · Trinity College, University of Toronto
Dean’s List Scholar · University of Toronto
Educational Training Stipend · Department of Mechanical & Industrial Engineering, University of Toronto
