American Hospital DubaiMedricsAI integration handbook

Operations and ownership

A clear boundary keeps AHD in control of conversations and Medrics accountable for gateway reliability, model configuration, and knowledge retrieval.

Sprinklr
Medrics
OpenAI
Handoff

Responsibility boundary

AreaAHDMedrics
Conversation and case recordOwn and retain in SprinklrNo application transcript store
Dialogue tree and queueConfigure, test, and approveProvide contract and support
Gateway runtimeConsume the HTTPS APIDeploy, monitor, patch, and recover
Bearer tokenStore securely and report exposureIssue, rotate, and revoke
AI prompt and knowledgeApprove AHD behavior and sourcesVersion, test, and operate
Production changeComplete AHD UATApply controlled release and rollback

Incident sequence

  1. Move affected conversations to the approved AHD human queue.
  2. Record UTC time, Sprinklr case ID, gateway request ID, status code, and frequency.
  3. Contact the agreed Medrics support channel without including credentials or unnecessary patient content.
  4. Medrics checks Container Apps health, revision status, gateway error events, and OpenAI availability.
  5. After recovery, run the synthetic smoke test and confirm dialogue routing before restoring automated traffic.

Change control

Model, AHD policy, vector-store, API contract, and bearer-token changes require a recorded owner, test evidence, release time, and rollback path. A new source-controlled policy release is tested before it reaches production. The previous approved container revision remains available for rollback.

Monitoring

OpenAI abuse monitoring is enabled for this project.

Go-live checklist

  • Production DNS and managed TLS certificate are active.
  • Sprinklr stores the bearer token outside visible dialogue variables.
  • All acceptance tests pass with synthetic data.
  • The human queue is staffed and receives the full Sprinklr context.
  • AHD and Medrics operations contacts acknowledge the handoff.