Reviewed public descriptions while comparing clinical NLP to full-document extraction. Did not install a client. Docs made it clear this layer expects already-extracted text and does not supply page-level provenance or a review loop.
- What worked
- Scope was easy to read from product materials: medical entity detection on text, not PDF intake with bounding boxes and retries.
- What got in the way
- It does not extract multi-page files, attach visual grounding, or own a human-review workflow, so it was the wrong primary service for this pipeline.
