AI Engineer & Data Scientist · Seattle, WA
I build ML systems in healthcare that ship to production and move real numbers, from $10M audit-classifier randomized controlled trials to agentic ChatGPT evaluated in the real world.
01What I do
I'm an AI Engineer & Data Scientist working at the seam between healthcare research and production. I take models the whole way, ETL, evaluation, and monitoring included, so they still move real numbers months after launch. Across three roles I've directed data science over a $31B revenue cycle, owned production ML at the largest publicly operated health plan in the U.S., and published two peer-reviewed transformer papers.
Director of Data Science
R1 RCM
Apr 2026 – Present
Data Scientist II → Applied AI Scientist III
L.A. Care Health Plan
Feb 2023 – Apr 2026
Machine Learning Researcher
Georgia Southern University
Sep 2021 – Dec 2022
At L.A. Care I ran a $10M audit-classifier randomized controlled trial on agentic ChatGPT-driven clinical NLP, real trials with real endpoints, not a notebook that scored well offline. I build the ETL, evaluation, and monitoring around the model so it still works months later.
An LLM-integrated medication NER system on Azure that caught regulatory risk in unstructured clinical notes and prevented $2M in violations, grounded, auditable, and wired into systems people already use.
02Selected work
All projects
An agentic AI bot that finds jobs, tailors every answer, and auto-applies, then tracks each application.
A multi-agent auto-application system: it navigates job sites, reads each posting, writes tailored cover letters and open-ended answers under hallucination control, scores experience fit, and submits, all logged to a live application tracker. (Company names in the screenshot are redacted.)
Web-mines and aggregates in-network care across Medi-Cal plans so anyone can find help they can actually reach.
Access Medi-Cal LA knocks down the barriers that keep people from care they are entitled to. It web-mines and aggregates provider and resource data across multiple Medi-Cal plans, reconciles it against L.A. Care's live FHIR directory, and surfaces in-network options a member can actually get to, ranked by distance or real travel time (walk, bike, drive) via Google Maps Platform APIs (Geocoding, Distance Matrix, Directions / Routes) on a Leaflet / OpenStreetMap map. Built for people without a car or an obvious place to start.
Assigning ICD codes from clinical text automatically, at 71% F1 on a notoriously long-tail label space.
A medical-coding model that maps unstructured clinical notes to ICD codes, reaching 71% F1 against a highly imbalanced, long-tailed label set. 🔧 Expand: the architecture, dataset, and how you handled the tail.
03Research
peer-reviewed publications
in averted adverse outcomes (RCT-validated)
audit RCT on ChatGPT-driven clinical NLP