Doctor Alliance
Turning multi-EHR clinical records into datasets a model could actually be trained on.
Live Project Intern, Tech
Doctor Alliance runs a HIPAA compliant platform for medical administration and revenue workflows. My half of it sat upstream of any model: pulling records out of two different EHR systems, normalising CPT codes, insurance details and CCN notes into one shape, and writing validation rules so payer and procedure data was clean before anything automated touched it. Alongside that, 40 to 50 clinical orders a day by hand, turning semi structured provider notes into standardised templates aligned with RCM logic. That is the least glamorous and the fastest way to learn what the data actually looks like.
What that meant day to day
- Normalised CPT, insurance and CCN notes across two EHR platforms into ML-ready datasets.
- Wrote validation rules that caught payer and procedure inconsistencies before automation.
- Processed 40 to 50 clinical orders a day into standardised, RCM aligned templates.
- Designed annotation patterns for entity extraction (procedure, payer, modifiers) to support later LLM prefill.
- Mapped RCM workflows to AI openings: auto-coding, anomaly detection, billing assistants.
What I’d point at
the annotation schema came out of doing the manual work first. you cannot design labels for a document type you have only read about, and I tried.