I turn complex operational data into systems people can trust.

Determining which routines were active, obsolete, or safe to retire would have required months of manual code review.

Built a Python framework to analyze Epic M routines and programmatically insert execution tracking across the codebase, then used production metrics to distinguish observed activity from dormant code.

Converted code discovery into a repeatable, data-driven process and gave modernization decisions a stronger evidence base.

Provider records arrived with inconsistent formatting and required a largely manual cleanup process before they could be imported reliably.

Developed a Python workflow to extract source data, apply repeatable normalization rules, and produce standardized records for downstream Epic administration.

Replaced repetitive cleanup with an auditable transformation pipeline and improved consistency across the import dataset.

A large portfolio of enterprise reports needed to move to a new delivery model without recreating each configuration by hand.

Built Python automation to generate report configurations, XML definitions, and batch email delivery workflows from structured inputs.

Significantly reduced manual migration effort while creating a more consistent and repeatable configuration process.

Assess whether a clinical laboratory outcome could be predicted from a mixed set of NHANES variables while maintaining a sound evaluation process.

Performed exploratory analysis, preprocessing, feature engineering, model comparison, and hyperparameter tuning in Python with scikit-learn.

Compared regression approaches—including Random Forest and Gradient Boosting—to identify the strongest-performing method and understand model tradeoffs.

Understanding LLM tradeoffs requires direct experience with inference, model behavior, prompt design, privacy, and deployment constraints.

Designed and maintained an Ollama-based environment to evaluate Mistral and other open-source models, experiment with prompts, and prototype AI-assisted development and automation workflows.

Built practical intuition for the differences between local and cloud AI—including capability, latency, control, and operational complexity.

Integrate a machine learning model into a responsive mobile camera workflow without relying on a remote inference service.

Combined TensorFlow Lite with Android’s Camera2 API to capture frames and run object detection directly on the device.

Demonstrated an end-to-end edge ML workflow with responsive inference, offline operation, and hands-on mobile model integration.