The first extraction pipeline used a provisional ophthalmology vocabulary across English, French, German, Portuguese, and Spanish. It looked for bounded mentions, laterality, negation, and selected high-risk terms instead of attempting a broad clinical interpretation.
Each resulting candidate retained the transcript text and timestamp range that produced it. The link made provenance inspectable, but it did not prove that the transcript or candidate was clinically correct.
A deterministic renderer placed candidates into fixed draft sections. Claim guardrails then compared draft wording with the available candidate evidence and produced issues when wording appeared unsupported or required attention.
This sequence fixed the intended order of authority. The transcript remained source material, extraction produced candidates, rendering produced a draft, and the physician remained responsible for review.
← Back to ClinScribe Local