Arun Mehta

Forty-one percent fewer manual classifications followed a browser-based review application I built for a research team. My academic work in human-computer interaction taught me to measure what automation changes, not just whether a model can produce an answer. I combine field interviews and task observation with evaluation sets, then implement accessible front ends, authenticated APIs, databases, and fallback queues for model-assisted workflows. Reviewer decisions feed monitoring dashboards so teams can detect drift and decide which steps should remain firmly human. Travel through rural clinics changed how I think about fragile connectivity and shared devices. I now test responsive pages, offline behavior, and synchronization under those constraints early, giving teams evidence for a release decision and a service their operators can understand.