Production ML in a Modern SaaS EHR

At Core Solutions Inc. I work on a SaaS EHR product being built from scratch, leveraging organizational learning from decades of operating a large-scale enterprise EHR platform.

My role is end-to-end on the ML side: I take ownership of complete product modules and develop full vertical slices that span ML model selection and data preparation, Python-based backend services, REST APIs, and the user-facing UI. Most features I own start as a product question — what the model is supposed to do in a clinician's workflow — before settling on an approach.

What the work looks like

• Modular ML components. Models are integrated as Python services behind thin API layers, designed so individual components can be swapped or upgraded without touching the rest of the product.
• Rapid AI prototyping. New AI-driven features are prototyped quickly and iterated on based on internal feedback, usage patterns, and evolving requirements.
• Evaluation workflows. Before features reach QA, they pass through evaluation pipelines that track model behavior across iterations to catch regressions and surface distribution shifts early.
• Cross-stack debugging. A surprising amount of work is debugging issues that span data, models, APIs, and UI integration — the bug is rarely cleanly in one layer.

Healthcare SaaS makes operational concerns (reliability, observability, the cost of getting things wrong) load-bearing rather than incidental. The job is less about novel models and more about being a dependable, debuggable component of a larger production surface.