Use when extracting structured reaction data (reactants, products, conditions, catalyst, solvent, T, t, yield) from a chemical reaction-scheme IMAGE — a cropped figure from a paper or patent, not a whole PDF. Wraps RxnScribe (Coley group, MIT — pix2seq scheme parser trained on USPTO patent schemes), which delegates molecule recognition to MolScribe and condition text to OCR. Output is JSON with role-typed bounding boxes and per-molecule SMILES. For whole-page PDF input, segment to scheme images first (PyMuPDF, pdf2image, or DECIMER-Segmentation). For single-molecule images use `mol-image-to-smiles`. Trigger on phrases like "extract reaction scheme", "parse reaction figure…
