The pipeline runs 5 Pydantic schemas in parallel per PDF chunk, each targeting a different clinical domain. Results are deep-merged into a single row per paper. Each of the 60+ fields includes a ...