About

I work in the part of a data stack nobody demos: moving records between systems, cleaning names that three departments spell three different ways, and deciding when a language model is the right tool and when it is an expensive way to guess. What I care about is the part after it ships. A pipeline that quietly did nothing at 3am looks exactly like a pipeline that worked, and most of what I build exists to tell those two apart.

At Telkom I inherited an overnight Pentaho job that kept failing without telling anyone and rebuilt it on Dagster and Airbyte, carrying around 10,000 transactions a day. A sensor asks the source one cheap question on an interval and only starts a sync when there is something to sync, and every run writes a row to a log table whether it moved data or not. Alongside it I built the same Indonesian place-name normalizer four separate ways, measured each on tokens, cost, runtime and whether a result could be traced back to its cause, and recommended the version that calls the model last rather than the one that scored best.

The rest ranges wider than one job title: a star-schema warehouse with 174 data-quality tests, three of them proven by deliberately breaking the pipeline to watch them fire; a two-factor campus gate for my thesis, where the face model's live error rate of 28.7% is published next to its lab figure of 1.59%; deployment pipelines built by hand on both Google Cloud and Azure. Information Systems at ITS, graduated August 2026, now looking for a data or AI engineering role.