As a software intern, I was actively involved in developing new features using Django and PostgreSQL on AWS, contributing to a robust and scalable backend. I participated in code reviews to maintain high standards of code quality and consistency across the team. My role included coding in Python for the backend and Node.js for the frontend, ensuring seamless integration and performance. Additionally, I researched new technologies and tools to enhance our platform, including potential AI projects for the future.

The aim of this project is to create a diffusion model that generates echocardiogram videos. Firstly, research was done to gain an understanding of the workings behind diffusion models. The mathematics and model architecture of these models were explored in depth from various academic papers for the means of learning to adapt them to our work. One of my main roles is to reproduce results of the paper Feature-conditioned Cascaded Video Diffusion Models for Precise Echocardiogram Synthesis. This work requires me to train an ejection fraction (EF) regression model and a diffusion model, then run an evaluation model for the diffusion and finally train the EF model on ablated and generated data. During this process, I have practiced debugging and have learned to understand the code behind these models.

This project is an inquiry for a cardiovascular diseases biomarker from Lung CT analysis. My role for this project was to quantitatively analyze the lungs with a software that identifies and segments the airways to create 3D models of the lungs. During the evaluation of 50 cardiac CT cases, I performed various troubleshooting techniques to resolve technical issues of the software. From this process I concluded that quantitative lung analysis is feasible within Cardiac CT field of view. At the end of my work term I delivered a presentation to a large audience of research professionals and doctors to share these findings.

• Conducted a literature review of LLM safety tooling across major providers (OpenAI, Google, Meta, Microsoft, Anthropic), identifying the shared gap between content-level moderation and explanation-quality evaluation
• Building a web platform to evaluate LLM reasoning faithfulness on toxic text datasets, addressing the gap between classification accuracy and explanation quality
• Planning semi-structured interviews with ML practitioners and research engineers to assess how LLM reasoning is currently used in AI safety workflows, informing the platform's design requirements

• Designed and built a club management platform serving 60+ active users using React.js,Node.js/Express and PostgreSQL.
• Integrated AWS S3 and Cloudfront for scalable image storage and persistence data management.
• Developed real-time workout signup and waitlist system with automated email notifications.
• Built an admin dashboard for member management, attendance tracking and analytics.
• View it here! https://www.uoft-tri.club/

• Partnered with Perimeter AI to tackle a class imbalance problem in their OCT scan cancer detection algorithm.
• Explored data augmentation strategies including conditional GANs, elastic transforms, and object detection techniques to synthetically expand the minority class.
• The augmented dataset produced a 5% improvement in test accuracy on ResNet-18 — a meaningful gain in a domain where false negatives have real consequences.

• Performed sentiment analysis on hotel review data scraped from Tripadvisor and plotted output to visualize results
• Designed a prompt engineering experiment to evaluate various prompt styles for a pretrained LLM on a task involving extracting specific information from email chains

Built this site from scratch by teaching myself HTML, CSS, and JavaScript — it was the project that kicked off everything else.

• Created an algorithm to perform kmeans clustering by accepting data of any size and programmed a cost function to find the optimal number of groups the datashould be separated into
• Created SVM and binary classifier algorithms in Python
• Programmed a Recurrent Neural Network to take as input a movie review and output whether it was a positive or negative review

• Built a self-balancing robot with a team of four for our open ended design project, treating the problem as an inverted pendulum control challenge
• Design decisions around height, centre of gravity, moment of inertia, and wheel sizing were grounded in physics theory
• The python control loop read IMU sensor data to estimate tilt angle and drove stepper motors to compensate in real time
• Our final product could stay upright for a record of 10 seconds!

Triathlon is a tough endurance sport that involves swimming, biking and running. With the team at the University of Toronto, I have consistently trained six times a week in the pool, gym and on the track for the past four years. This season I started my role as Co-President to share and promote a shared passion for the sport! I have now competed in many races, including the amateur World Championships last year in Spain and this summer I am planning on racing my first Olympic Triathlon!

This summer I spent many nights alone in the wilderness. I slept in hammocks, tents, the back of a car, and even just in a sleeping bag under the stars. Before setting up camp I would hike 10-20 kilometers with a 40lb backpack through forests, up mountains and along lakes. These adventures pushed me to extend the limits of my physical and mental boundaries to the point where I am now comfortable taking on expeditions on my own.