Waterloo, Ontario, Canada
University of Waterloo Computer Science undergraduate student. Interest in Machine Learning and Software Development. Eager to learn more and gain valuable work and research experience.
Selected for the NSERC USRA and conducting Machine Learning research with the Vector Institute. Co-first author of "Work While They Sleep: Exploiting Evaluation Latency for Fully Bayesian Optimization," submitted to ICLR 2027. Training transformers via reinforcement learning to perform multi-fidelity Bayesian optimization on function networks. Researching advanced reinforcement learning strategies for earth element recycling.
Researched Bayesian optimization and machine learning techniques for catalyst design in CO2 conversion. Implemented a model-based reinforcement learning framework using PPO for non-myopic Bayesian optimization. Explored non-myopic Bayesian optimization using low-fidelity rollouts to address model misspecification.
Worked on Software Factory, creating agentic systems to redefine software creation. Built custom systems around AI agents to improve reliability and performance in agentic coding workflows. Developed 3 full-featured modules for a full-cycle SDLC platform, covering requirements tracking, project management, and QA workflows, each delivering complete end-to-end functionality across multiple features and user flows. Contributed across product and engineering by defining product requirements, writing high-level engineering blueprints, building full end-to-end features, and shipping improvements throughout the codebase.
Co-authored “DermArt: A Human-Annotated Dataset and Contrastive Distillation Framework for Artifact-Robust Dermatological AI,” submitted to MICCAI 2026 (second author). Contributed to developing a contrastive distillation framework integrating contrastive representation learning and self-distillation to improve skin cancer diagnosis robustness and reduce disparity between clean and artifact images. Benchmarked CLIP-based vision-language models and multimodal LLMs for zero-shot artifact annotation on the Fitzpatrick 17k dermatological dataset, evaluating model robustness across diverse artifact types.
Designed and trained deep learning models in PyTorch to decode typed text from EEG data, leveraging convolutional neural networks (CNNs) and attention mechanisms to reconstruct keystrokes from brainwave signals. Engineered end-to-end EEG processing pipelines, including signal preprocessing, temporal windowing, and feature extraction.
Led projects and collaborated directly with customers to provide tailored solutions to complex technical challenges. Built an AI RAG chatbot that retrieves and ranks context, delivering accurate answers on internal design documents. Built a self-improving accounting reconciliation system and a part request application with dynamic auto-approval.