Jason Li

I work on computer vision models for analyzing zebrafish behavior, using Python, OpenCV, and YOLO to process video data and identify patterns. The challenge isn't just building accurate models. It's understanding what behaviors matter to neuroscientists and how computational methods can answer questions that traditional analysis can't. Real data is messy, which is part of what makes it interesting. This work has taught me how to bridge the gap between technical implementation and biological understanding, often working with datasets that require careful preprocessing and domain expertise.

I developed transformer models for analyzing large-scale social media data, focusing on extracting meaningful patterns from complex text datasets. This project resulted in a paper published at the Second International Conference on Informatics (ICI-2023) and gave me experience working with state-of-the-art NLP methods while thinking critically about how these models can be applied to real-world problems. The work required balancing technical innovation with practical constraints, and it reinforced my interest in applying AI to domains where the data is inherently complex and meaningful.

I worked with large patient datasets, performing statistical analysis and building Shiny visualizations to help researchers understand patterns in health data. This experience taught me how to think about data quality, statistical rigor, and visualization design, all while working with datasets where the stakes are high and the questions are complex. The work required close collaboration with domain experts to ensure that the analysis and visualizations actually supported research goals.

I work on perception systems for UAVs using PX4 and ROS2, developing computer vision pipelines that enable autonomous navigation. The challenge is building systems that work reliably in real environments, not just in simulation. This has taught me systems thinking, how to debug complex autonomy stacks, and how to balance theoretical approaches with practical constraints. The work requires understanding everything from low-level sensor data to high-level planning, and it's given me experience in how software interfaces with physical systems.

I developed features in React Native for a mental health application used by over 2,000 users. This work taught me how to think about user experience, iterate based on feedback, and ship software that people actually use. The product sits at the intersection of technology and health, which reinforced my interest in building tools that can have real impact. Working on a deployed product with real users forced me to think about edge cases, performance, and how technical decisions affect user experience.