jugmt is a minimalistic spex (SPecification EXtractor) implementation
with a codebase less than 200 lines of Python. The tool extracts figure
information and tables from .docx files, generates HTML and JSON,
and validates the JSON using a JSON schema.
When running the tool on a collection of NVMe specification documents, including Base, Boot, MI, NVM, ZNS, KV, PCI, RDMA, and TCP, it consumes a total of 5 seconds of wall-clock time and about 500MB of memory on an i7-1360P using a single thread for all documents combined.
For more information on the source code, extracted table formats, and validation, please refer to the online documentation