Can maps be made accessible? Rethinking data visualisation for screen readers

Most data visualisations are fundamentally visual — they rely on colour, position, and shape to convey meaning. But what happens when a user can't see the chart? This essay documents my semester-long project building an accessibility layer for SVG-based campus maps, and what I learned about the limits of alt text.

Predicting Seoul subway delays: a machine learning retrospective

I spent a semester training a delay prediction model on three years of Seoul subway data. The model works. More interestingly, the ways it fails reveal something about the nature of urban transit and why "on time" is a political concept as much as a logistical one.

What writing a tiny interpreter taught me about language design

For my Programming Languages course I built a small interpreter from scratch — lexer, parser, evaluator, REPL and all. This is less a technical write-up and more a reflection on how building something forces you to form opinions you didn't know you had.

Notes on fine-tuning large language models efficiently

A write-up from my time as a visiting student at the KAIST ML Group. We looked at parameter-efficient fine-tuning methods — LoRA, prefix tuning, adapters — and tried to understand when each approach actually makes sense.