Demo environment. Do not enter real Protected Health Information (PHI). Use synthetic data only until production BAA infrastructure is wired.

How it works

The Centaur loop

AI drafts. You verify. The system signs. No step skips the human, and no step skips the audit.

01

Capture raw context

Paste shift notes, dictate observations, or import structured vitals. Before anything is sent to a model, a redaction pass strips names, DOBs, MRN-like patterns, SSNs, phone numbers, and ambiguous date strings. The redacted version is shown to you for review.

REDACTED, DOB REDACTED [PATIENT_A] [DOB_REDACTED]

02

AI synthesis

The redacted context is sent to an enterprise LLM with a strict nursing-documentation system prompt. Output formats: SBAR, narrative, focus note, handoff. The provider is wrapped behind a single interface, so swapping to Azure OpenAI / AWS Bedrock under your own BAA is one config change.

03

The diff view

AI-generated tokens are highlighted in light blue. The moment you edit a span, it turns green. You see what is AI vs. what is human before you sign anything.

Patient reports 7/10 substernal chest pain radiating to the left arm, onset approximately 30 minutes prior to assessment.

04

Approve & Sign

The Copy and Export buttons stay greyed out until you tick the attestation checkbox and click Approve & Sign. On click, the server computes a SHA-256 hash binding the AI output, your final text, your verified license, your user ID, IP, and timestamp. The chart locks. The audit log gets an immutable row.

hash: f81d4fae-7dec-11d0-a765-00a0c91e6bf6 …