I build AI products that solve real business problems.

Many payer workflows involve reviewing medical records against different criteria. Long records and information spread across pages make it difficult to find the evidence each review requires. I designed a reusable product around that shared need, so reviewers can provide medical records and review instructions to get findings tailored to their task.

I conceived and built the product, led its AI architecture and evaluation approach, and partnered with engineering on the interface and delivery. It extracts evidence and connects findings to the source medical records.

Value to the teamReviewers receive a tailored summary and extracted findings linked to supporting passages, so they can locate pertinent evidence and verify it in the source medical record.

Risk adjustment accounts for differences in members’ health needs when determining payments to health insurers. Reviewers need complete, accurate diagnosis information tied to the relevant date of service, with supporting evidence that can be verified in the medical record.

I designed and built an AI pipeline that identifies potentially risk-adjustable diagnoses, links them to visit dates and supporting evidence, and supports both targeted chart review and more efficient coding. I led the architecture, implementation, and evaluation end to end.

Value to the teamChart prioritization focuses human coding effort on records with potentially missed diagnoses. Coding support gives reviewers findings and source evidence they can inspect efficiently, while coders retain responsibility for coding decisions.

Teams reviewing scanned medical records often have to read page by page to locate relevant information. A searchable record can remove a substantial amount of that navigation work.

I led architecture and development, partnered with engineers on deployment, and worked with product and healthcare leaders to understand review needs and support adoption.

Value to the teamMakes pertinent evidence easier to find across workflows such as medical necessity review, claims adjudication, and DRG review.

Health plan quality reviews need clinical evidence that may be missing from claims data. Reviewers examine medical records to locate measurements and documented care relevant to quality measures.

I developed models for diabetes-related evidence, including blood pressure, A1C, and eye exams. I also supported annotators and worked through model-training issues as the team modernized its training tooling.

Value to the teamHelps reviewers identify evidence relevant to care gaps and complete medical-record review more efficiently.

As more teams build AI products, technical review needs to keep pace while remaining consistent and useful to the people delivering those systems.

My co-chair and I lead a team of data scientist reviewers responsible for technical review of AI use cases across the enterprise. I also develop guidance on evaluation and responsible AI development. The scope spans traditional machine learning, generative AI, retrieval-augmented generation, and agentic systems.

Value to the teamHelps teams make informed technical decisions and build stronger evidence for the quality of their AI products.