Trust
Built for clinical responsibility
Think of how exam papers get graded by roll number, not by name — the grader sees the answers, never whose they are. Ritaja Practice’s AI works the same way: before anything reaches the model, every identifying detail is stripped out — no patient name, no clinician name, no practice name, nothing that could tell one client’s story from another’s. That holds even in group and couples sessions with several people in one room, not just one-to-one care. What comes back is a draft, not a record, and the last word is never the AI’s — every draft is reviewed and signed by the clinician treating that client, or edited first.
Runs in our own Azure tenant
AI requests go to Azure OpenAI inside our own tenant under a Microsoft Business Associate Agreement — not the public OpenAI API, and not a third-party AI vendor. Transcription uses Azure Speech in the same tenant, with no outside scribe.
Identifiers do not travel
Chart assistants carry no name, date of birth, or contact field. Free-text notes pass through a PHI scrubber before a request leaves the chart.
Never auto-triggered
A note only drafts after a real workflow action — a session, a scratchpad entry, a chart question. Staff and patients never get AI output without you taking that step.
Covered by a signed BAA
Writing assist and optional transcription are both covered under a signed Business Associate Agreement when configured for HIPAA workloads.
Metadata audit, not note bodies
We log that an AI request occurred — never the full text of clinical notes or prompts in routine audit trails.
Clinician in the loop
No auto-sign, no auto-send to portal, no hidden recording. Transcripts and AI drafts are scratchpad material until you approve.