Writing on
applied ML.

Long-form writing for the technically curious — on research methodology, clinical data engineering, and the realities of building ML systems for medicine.

01
Research CraftNovember 15, 202412 min read

What Makes a Good Ablation Study

Ablation studies are the backbone of rigorous ML research, but most papers get them wrong. Here's how to design experiments that actually answer the right questions — and convince reviewers you know what you're doing.

Machine LearningResearch MethodsDeep Learning
02
ScienceOctober 22, 20249 min read

ECG to Glucose: The Physics of the Problem

Why would your heartbeat tell you anything about your blood sugar? The surprising physiological mechanisms linking cardiac electrical activity to glycemic state — and why this relationship is harder to model than it looks.

ECGGlucosePhysiologyBiosignals
03
EngineeringSeptember 30, 202415 min read

The Researcher's Survival Guide to MIMIC-IV

MIMIC-IV is one of the most powerful clinical datasets available — and one of the most frustrating. BigQuery schemas, temporal alignment pitfalls, pharmacological confounding, and the leakage traps nobody warns you about.

MIMIC-IVClinical DataBigQueryData Engineering
04
EngineeringAugust 18, 20248 min read

Edge AI for Medical Devices: Constraints Are Features

Deploying ML on ESP32 and embedded hardware sounds like a downgrade. In medical AI, the constraints of edge deployment — low power, no cloud dependency, sub-10ms latency — are exactly what make a device trustworthy.

Edge AIEmbedded SystemsESP32Medical Devices