Senior Product Manager · Healthcare Data & Analytics

I turn requirements into working, interactive products — fast.

Product manager for data and analytics products — business-first, customer-in, and technically hands-on. I turn requirements into interactive, testable interfaces that align stakeholders and de-risk the build before engineering commits.

See the work How I work

Everything below is live and clickable — try the demos directly on this page.

Selected work

Each project follows the same arc: a real product problem, an interactive prototype built with AI code assistants, and a measurable outcome.

01 · AI feasibility analysis

Feasibility Cohort Explorer

Ask for a patient cohort in plain English, get an instant validated count across 114,594 synthetic patients.

Problem

Every clinical study starts with "are there enough patients who meet these criteria?" Answering means a data analyst hand-writing SQL, so each question becomes a ticket and a wait. Pointing an LLM at the database instead trades the wait for a worse problem: confident-looking queries that reference codes that don't exist.

What I built

A chat-driven cohort builder with live counts and demographic breakdowns. The LLM's only job is producing a small, validated criteria structure grounded to real ICD-10-CM, NDC, and LOINC codes — a deterministic, LLM-free compiler turns it into parameterized SQL. The same engine also answers over MCP.

Outcome

  • Zero hallucinated codes by construction — every code validated against 261k+ vocabulary entries before execution
  • ~75 ms median cohort counts across 114,594 patients
  • One engine, two surfaces — web app and MCP server return identical counts
Feasibility Cohort Explorer interface: chat pane with example queries like 'Diabetic patients over 60 on metformin' beside a live cohort summary panel
Open full screen ↗

02 · Feasibility tooling

Codeset Builder

Search, browse, and assemble medical code sets across 261k+ ICD-10-CM, HCPCS/CPT, and NDC codes.

Problem

Scoping a data feasibility request starts with defining the clinical codes involved. Doing that across hundreds of thousands of hierarchical codes with spreadsheets and email threads was slow, error-prone, and hard for non-clinical stakeholders to follow.

What I built

An interactive code search and selection tool — hierarchical browsing, cross-code-system search, and CSV import/export — prototyped end-to-end with AI code assistants so stakeholders could react to a working interface instead of a spec.

Outcome

  • 40+ hours of product design time saved — components shipped directly into the final product
  • $5M+ enterprise partnership supported by the feasibility workflow it prototyped
Codeset Builder interface: searchable, hierarchical list of ICD-10-CM codes with a selected-codes panel and CSV upload
Open full screen ↗

03 · Clinical data exploration

Unstructured EHR Search

Boolean query interface for exploring de-identified, unstructured electronic health records.

Problem

Most of the signal in electronic health records lives in unstructured notes. Customers needed a way to test whether specific clinical patterns existed in the data — without writing queries or waiting on an analyst.

What I built

A search interface with composable INCLUDING / OR / EXCLUDING operators over de-identified records, so a non-technical user can express a real clinical question in seconds. Prototyped with AI code assistants against fully synthetic data.

Outcome

  • Prototype components carried directly into the production build
  • Gave stakeholders a working interface to react to before engineering committed a line of code
EHR Search interface: search bar with INCLUDING, OR, and EXCLUDING boolean operator chips for querying de-identified records
Open full screen ↗

All demos use synthetic or public data only.

How I work

Prototyping is the method, not the identity — it's how I make product decisions cheaper and faster.

  1. 1

    Start from the problem

    Customer conversations and business requirements first. I frame the decision the prototype needs to de-risk before anything gets built.

  2. 2

    Prototype in days, not quarters

    Using AI code assistants (Cursor, Claude Code), I turn the requirements into an interactive, testable frontend that customers and stakeholders can actually click.

  3. 3

    De-risk, then build for real

    The prototype aligns stakeholders and surfaces the hard questions early. Then I partner with engineering on the production build — often carrying components straight across.

About

I'm a senior product manager working on data and analytics products in healthcare at Inovalon. Before that, I was a consultant at Deloitte. I studied Chemical Engineering with a Computer Science minor at Georgia Tech — which is roughly the ratio of business to hands-on-technical I still operate at today.