Read what we actually write.

The outage lasted eleven minutes. A recommendation service I helped maintain went down during a regional sales event, and for eleven minutes the platform served every customer the same static list of products. What stayed with me was not the incident itself but the postmortem: nobody on the team, myself included, could explain precisely why the model had degraded in the weeks before the failure. We could read the code. We could not read the system. I decided that week that I wanted to become an engineer who could.

For the past three years I have worked as a backend engineer at a logistics company in Jakarta, where I own the services that route roughly forty thousand shipments a day. My strongest work has been at the seam between infrastructure and machine learning: I rebuilt our ETA-prediction pipeline after inheriting it from a departed data scientist, cutting median prediction error by a third — not by changing the model, but by discovering that a third of its training data carried timestamps from the wrong time zone. That project taught me a lesson my degree never did: in production systems, the model is rarely the problem. The system around it is.

I keep meeting the limits of what I can teach myself. I can deploy a model, monitor it, and roll it back; I cannot yet reason formally about why distribution shift breaks it, or design training procedures that anticipate failure instead of reacting to it. Papers on ML reliability assume mathematics I have circled but never studied deeply — optimization, statistical learning theory, causal inference. Self-study fills gaps; it does not build foundations.

That is what draws me to a research-oriented M.S. rather than more years of production experience. I want the coursework in machine learning systems and the chance to do supervised research on reliability — how models fail quietly, how monitoring can catch semantic drift before customers do, and how engineering practice should change when software behavior is learned rather than written. My background means I arrive with questions from real systems, not just curiosity: I have watched the failure modes I want to study.

After the degree, I intend to work as a machine-learning infrastructure engineer on a team that treats reliability as a first-class problem, and eventually to lead one. The engineers I admired during that eleven-minute outage were the ones who could move between a stack trace and a statistics question without changing gears. This degree is how I become one of them.