DField SolutionsMérnöki stúdió · Budapest
Loading · Töltődik
Skip to content
CASE STUDIES · health

AI support copilot for e-health SaaS · 38% deflection, 0 PHI leaks in 6 months

A Hungarian e-health SaaS needed AI-powered support that never leaks PHI and always cites a source. We shipped a Claude Opus + pgvector copilot with strict refusal guardrails; 6 months in, zero PHI incidents.

Timeline11 weeks
← Back to case studies
Reviewed by
0138% of tier-1 tickets deflected in month 3.
020 PHI leakage incidents in 6 months (validator caught 14 attempts).
03 / Average CSAT on AI-answered tickets4.4/5.
04CMO-approved citation pattern became a company-wide standard.
The problem01 / 03
  • 01Tier-1 ticket volume grew faster than hiring could keep up.
  • 02PHI exposure was a company-killing risk · needed hard guardrails.
  • 03Every answer had to carry a citation to pass the CMO review.
  • 04FHIR-formatted patient data couldn't be put in a general-purpose vector index.
The solution02 / 03
  • 01Claude Opus for strict refusal behaviour on medical / PHI content.
  • 02FHIR-aware retrieval with tenant-scoped pgvector index + provenance chain.
  • 03Output validator reverse-checks for PHI leakage before returning.
  • 04Citation mandate · no answer without a source-doc URL visible to the user.
The outcome03 / 03
  • 0138% of tier-1 tickets deflected in month 3.
  • 020 PHI leakage incidents in 6 months (validator caught 14 attempts).
  • 03Average CSAT on AI-answered tickets: 4.4/5.
  • 04CMO-approved citation pattern became a company-wide standard.
CASE STUDIES

Let's get started.

Send an email or book a 30-minute call.

Back to case studies