I use this blog for research notes, project build logs, experiment writeups, and short technical reflections. It is a lighter-weight space than the portfolio and research pages, so I can document progress, lessons learned, and small ideas as they happen.
Recent posts are shown below. You can also browse everything by year or by tag.
One of the most satisfying parts of my itinerary optimization project has been watching it evolve from a single-day attraction-routing prototype into a multi-day planning system with hotel selection, traveler profiles, and route visualization. The original version already handled utility, waiting time, and travel cost, but it still assumed the trip began and ended within one day.
In the PDEBench-Lang project, the main question is surprisingly simple: does the format of a partial differential equation change how well a language model can reason about it? We often focus on model architecture and training data size, but representation format can matter just as much when the task is symbolic.
In facial expression recognition, it is easy to stop at accuracy and say a model works. I have become more interested in a second question: what visual evidence is the model actually using when it predicts an emotion?