Resume

Social-care platform connecting health plans, providers and 100+ community organizations.

• Configured a customer’s member data model against a health plan’s file specification: 355 fields, per-field visibility and ownership. After watching intake at three organizations, replaced the one-off code with a shared configuration model: onboarding ~3 weeks to 4 days.
• Took a health plan through their security review and connected under their rules: federated SSO, allowlisted member-data exchange, key rotation. Cleared in 9 weeks with 2 findings I designed around rather than waived; a second engineer used my runbook for the next plan.
• Built the provider directory behind the referral engine: CMS NPPES, HRSA and IRS filings.
• Built the semantic matching layer over it, on sentence-transformer embeddings. Still in production.
• Built and owned the reply-intent classification pipeline in Python, PostgreSQL and Claude Haiku over 118.2K reply events across 15 intent categories I defined. The per-slice evaluation gate blocked 2 changes that helped the overall number but degraded non-English replies.
• Platform outcomes these systems support: 19.2% engagement across a ~955.6K-member Medicaid cohort (~183K engaged members, ~12.8x the published 1.5% CHCS SMS benchmark) and 35K+ client-reported enrollments and re-enrollments further down the funnel.

AI workforce for value-based care; FQHC and ACO customers.

• Deployed into an FQHC customer’s environment under three rounds of credential and network approval, debugging nine cross-boundary failures and confirming two in our implementation, escalating them with reproductions.
• Shipped 11 changes into an ingestion codebase I did not write, 4 revised after review by the engineer who owns that surface, mapping the customer’s EMR extract into our schema and resolving 23 field and format mismatches before first clean load.

• Investigated pharmaceutical fraud across clinical and public Medicaid data, making the signals interpretable to operational and finance stakeholders.