I am drawn to projects where the hard part is not only choosing a model. Forecasting sparse demand, for example, forced me to think about validation, zero-heavy targets, recursive prediction, and whether the features would actually exist at prediction time.

On the computer vision side, I have been experimenting with visual odometry ideas on drone footage. On the product side, I have worked with Flutter, Firebase, and speech or story-generation flows. Some of it is research-like, some of it is product-like, but I try to keep the work concrete.

I care about leaving a trail: what changed, what improved, what failed, and what I would do differently. That is why the projects here are written as case studies instead of only screenshots and tags.