Case 11 · Tier-3 Hospital Public/Private Hybrid Routing · Total Cost -46% · Zero PHI Egress

TL;DR: Tier-3 Hospital Public/Private Hybrid Routing · Total Cost -46% · Zero PHI Egress. Direction: Multi-Model Routing. Scale: tier-3 hospital. Hard KPI: total cost -46% vs full private, no PHI egress. Tech: PHI detection + DLP at gateway, on-prem AW36-72B, public GPT-4o for non-sensitive. My role: routing architecture, PHI detection rules, compliance assessment. Client name anonymized under NDA.

Problem

Full on-prem GPU is expensive, but hospital compliance forbids all requests going public. Sensitivity-based split needed.

Solution

Sensitivity-tier routing — traffic with PHI → internal AW36-72B, non-PHI (announcement drafting / schedule explanation) → public GPT-4o / Claude, gateway PHI detection + DLP.

Tech Stack

PHI detection + DLP at gateway, on-prem AW36-72B, public GPT-4o for non-sensitive

Result (Hard KPIs)

total cost -46% vs full private, no PHI egress

My Role

routing architecture, PHI detection rules, compliance assessment

Next: Download PDF · 30-min alignment

Client name anonymized · KPIs verified · Real client name disclosable under NDA.