Led 4 engineers to build the intake platform that became the foundation of Enzo's EHR, replacing manual workflows with an end-to-end system that improved onboarding speed, accuracy, and visibility for clinical and leadership teams.
Built a clinical document pipeline that processes messy referral packets (faxes, handwritten notes, inconsistent formats) through two stages: a third-party OCR/extraction service for structured data, then an LLM-powered reconciliation layer that resolves conflicting and outdated information — deduplicating medication lists against a drug database, synthesizing a patient story with diagnosis codes from raw clinical notes, and filtering stale records to produce a current patient profile. Validated accuracy through human review workflows and iterative refinement from clinical team feedback.
Architected Enzo's workflow engine after recognizing that AI extraction alone wasn't enough — clinicians still had to manually transfer results into their EHR. Built a DAG-based orchestration system with an event-driven executor: referrals arrive from integrated sources (referral portals, fax, email), trigger the document pipeline, and flow through a graph of automated and human-in-the-loop nodes — with automatic fallbacks that degrade gracefully (e.g., failed insurance verification spawns a human task with the UI context needed to complete it manually). Once a patient is accepted, the workflow builds the patient record directly in the customer's EHR. Built a drag-and-drop workflow configurator so a technical team can onboard new customers with zero code changes, wiring together node types with typed inputs and outputs to match each customer's intake operations.