About

I'm Jagnyesh Patra. I work at the Tufts Clinical and Translational Science Institute, where I evaluate AI tools for development and data discovery. I also built a self-paced trainer that takes SQL analysts through OMOP data. OMOP is the common data model that research networks use to put EHR data from different hospitals into one shape.

For nine years I have worked where clinical documentation and payer data meet the systems that run on them. That work has been at providers, payers, a pharmacy benefit manager, and the CDC. At Optum I set the mapping standards for turning HL7 v2 messages into FHIR resources, and 8 business units adopted them. HL7 v2 is the older message format most hospital systems still send. FHIR is the current HL7 standard for exchanging health records. I also led development of Patient Access APIs certified for USCDI v1, the federal list of data elements that certified health IT must exchange.

At the CDC I worked on federal and state data interoperability during COVID-19 and contributed to USCDI data element proposals. At the University of Illinois Chicago I designed an OMOP data layer that put data from several EHR systems into one standard representation.

From product to building

Most of that work was in product and analyst roles. I interviewed clinicians and principal investigators, wrote requirements, prioritized backlogs, and launched MVPs. I still start there, and now I write the code and the evaluations as well. The projects on this site are where I do that in public, on synthetic data only.

How I work

• I measure a claim against a baseline and publish the result when it goes against me. Inrcm-ai, the plain-LLM baseline beat my knowledge graph on accuracy, and the README says so first.
• I keep every fact traceable to the source row it came from.
• I put human review where a wrong answer costs the most.

Elsewhere

My résumé has the full work history. The code is on GitHub, and I'm on LinkedIn.