Clinical decisions need a path to evidence.

The real problem.

Documentation, records and medication safety often live in separate workflows. A useful system must connect them without turning persuasive language into an unverifiable clinical decision.

The architecture decision.

Separate probabilistic language tasks from the rule engine. Transcription and extraction produce structured data; contracts and a safety gate define what can be evaluated. The physician retains judgement and signs the decision.

The controls that support it.

• Data contracts that define the scope of 21 clinical modules.
• Twenty universal red flags before routine routing.
• Layered prescription validation: presentation, completeness, dose and relationship to diagnosis.
• A 259-file SHA-256 manifest with an integrity self-test.

Evidence with context.

Pilot 0 reports documentation time falling from 60 to 15 minutes per patient across 100 consultations. The CV also describes an E2E harness covering 45 screens. These measurements concern the documented scope and do not guarantee clinical outcomes.

Source for figures and scope: the author’s 2026 CV. This is not live telemetry.

The limits matter too.

Reproducibility alone does not demonstrate clinical safety. Rules, sources, inputs and professional review require validation for each context. This case is an engineering description, not a diagnostic tool.