I'm interested in solving transportation problems at the intersection of engineering, analytics, and technology.
I'm finishing my civil engineering degree at the University of Toronto in 2027. Over a year at the City of Toronto, I worked in the Data & Analytics Unit, across data collection, operations, and analytics. Projects ranged from World Cup congestion monitoring and city-wide bicycle demand estimation to accessibility modeling, machine learning, and Automated Speed Enforcement evaluation. This work supported operational planning, informed City Council reporting, and contributed to public transportation policy discussions.
What I enjoy most are problems where the answer isn't obvious from the data alone: understanding congestion, improving transit reliability, estimating travel demand, or evaluating policy. I like combining engineering judgment with analytical tools to help cities make better decisions.
I also like building under a deadline. I won the ShiftED hackathon with an AI patient-monitoring platform and placed 10th of 328 at the NSRI Summer Research Hackathon.
Outside of engineering, I'm usually building side projects, exploring new tech, or playing sports.
Engineering solutions for transportation, infrastructure, and healthcare.
Designed and deployed an automated system evaluating the impacts of ASE camera removals across Toronto. Built scalable analytics pipelines on city-wide traffic data to monitor roadway performance and support policy evaluation. The analyses informed City Council reporting and were discussed publicly by the Mayor.
Developed automated congestion monitoring workflows supporting the City of Toronto's transportation planning for the World Cup. Built repeatable Python and SQL pipelines to monitor network performance, identify congestion trends, and generate operational insights.
A city-scale pipeline for Annual Average Daily Bicycle Traffic (AADT), combining temporal normalization, seasonal adjustment, and confidence interval estimation to support cycling infrastructure planning.
Reproducible geospatial workflows computing gravity-based and cumulative accessibility measures across transportation networks, with GIS visualizations supporting planning and accessibility analysis.
Analyzed over ten years of Toronto streetcar data to identify delay patterns across multiple routes. Built and validated Linear Regression and Random Forest models in Python (Pandas, scikit-learn) to predict delay durations, with Random Forest outperforming on complex, non-linear relationships. Produced a route-level performance report with insights for scheduling and service optimization.
Read the write-up ↗Studied a Toronto streetcar route over three months, collecting 150+ multi-point data entries to inform operational improvements. Processed the data with Python and Pandas to produce visual reports on traffic flow and scheduling patterns, and published the findings on Medium.
Read the write-up ↗Placed 10th out of 328 participants (100+ teams) at the NSRI Summer Research Hackathon. Tested AI's likely effect on hospital costs against two historical analogies, slow post-electrification productivity gains vs. the dot-com boom-and-bust, using Medicare Cost Report data across ~6,000 hospitals (2011–2024). Found administrative cost share has climbed steadily since EHR saturation with no reversal, forecasted to keep rising through 2034, while clinical AI funding already shows classic bubble markers.
Read the report ↗Winner of the Translational AIxBio Hackathon. An AI-powered patient-monitoring platform that learns individualized physiological baselines from wearables to detect early deterioration. Currently being developed as a startup.
A proof of concept that predicts how two friends would rank Toronto restaurants from the way Yelp reviewers describe them. It embeds 37,509 review sentences, groups them into 166 topics, and fits a ridge regression per person on 24 aspect features, then combines both predictions into group picks. Held-out accuracy is 0.26 and 0.38 (Spearman) against 0.17 and 0.20 for random guessing, confirmed with a permutation test. Built on two people's lists as a test of the approach, so the sample is small.
Open the interactive app ↗ View the code ↗Led the analysis and modelling for a team project that ranks 29,868 km of Alberta highway by wildlife-vehicle collision risk, to show where crossing structures would prevent the most collisions. Built a reproducible 13-stage pipeline that cuts the network into 15,000 segments, snaps 52,000 carcass records onto them, and joins road, traffic, habitat, terrain and animal-abundance data. A negative binomial model scores every segment, validated on highways and years it never saw: the top 10% of segments capture 48.5% of observed records, nearly five times random. Benchmarked against boosted trees, hurdle, LDA and logistic models, and checked for data leakage. Results feed an interactive risk map.
Open the interactive risk map ↗ See the pipeline diagram ↗CGPA 3.50 · Dean's List ×3