Menu
Back to Case Studies

Healthcare ·Confidential, US healthcare provider

Clinical and RCM staff get their answer on the page they are already on, not five screens later.

A US healthcare provider whose support, RCM and clinical staff needed patient, claim and eligibility data fast, without loosening a single HIPAA control.

Where this fits

Best for support and operations teams in regulated healthcare that need staff to move through patient and claim data faster, without ever loosening HIPAA control.

1.5+ years live on a real clinical workload. Inside HIPAA and SOC 2 Type II.

  • HIPAA + SOC 2 Type II
  • Role-based PHI access
  • Answers grounded in records, not generated

What we did

  • Built a side panel assistant that reads the page the user is on
  • Wired the model to patient, claim and eligibility data through MCP
  • Enforced role-based PHI access at the tool layer, under the model
  • Shipped inside the existing compliant environment, no new data path

Result

Staff work in one place and ask in plain language instead of moving between internal screens. The assistant summarises a chart, flags what a payer will be missing, and drafts the response in place. The client reports roughly 40% less handling time per support ticket.

Side panel reading a live chart: note summary, missing documentation flagged against the payer's prior authorization rule, and the letter drafted in place. Synthetic data.

The hard part

An assistant over PHI cannot be secured in the prompt. Every answer it gives is a disclosure event, so role limits and audit logging have to sit in the tool layer beneath the model, and every fact has to come from a record rather than from generation. A model that invents a claim status is not a UX bug in RCM.

Stack
ReactModel Context ProtocolOpenAIClaude APIRetool
Focus
Healthcare AIHIPAASOC 2AI AssistantMCPPHI Access Control

Facing a similar problem?

Tell us where it stalls today, and we'll tell you how we would approach it.