⌘K
89% complete
34 reference numerals
~13 pages
Jurisdiction
United States (USPTO)
Type
Utility
Entity Size
Small
Technology
This invention relates to document processing and artificial intelligence, and more particularly to a system and method for transforming unstructured portable document format (PDF) files into self-contained, semantically enriched container files optimized for consumption by autonomous AI agents.
Prior Art
25 references found
A computer-implemented method for converting an unstructured electronic document into an Agent Document Format container file comprising receiving, by a document conversion system, an input unstructured electronic document; extracting, from the input unstructured electronic document, extracted text and positional metadata and typographic metadata associated with the extracted text; detecting, based on the extracted text and the positional metadata and the typographic metadata, a plurality of structural elements to generate structured document data, the plurality of structural elements including headings, sections, tables, and references; partitioning, based on the structured document data, the extracted text into a plurality of semantic chunks; generating, using a large language model operating on the plurality of semantic chunks, semantic enrichment data comprising a summary, a set of entities, a document classification, and a knowledge graph; generating, for each semantic chunk of the plurality of semantic chunks, a vector embedding; and packaging, into a single compressed archive file as the Agent Document Format container file, the input unstructured electronic document, the extracted text, the positional metadata, the typographic metadata, the structured document data, the semantic enrichment data, and the vector embedding for each semantic chunk, wherein the Agent Document Format container file is configured to enable an artificial intelligence agent to parse, query, and reason over the input unstructured electronic document using the packaged contents without re-performing extraction, structural detection, semantic enrichment, or vector embedding generation.
The computer-implemented method for converting an unstructured electronic document into an Agent Document Format container file comprising wherein detecting the plurality of structural elements comprises applying a set of heuristic rules to the extracted text and the positional metadata and the typographic metadata to assign, for each structural element of the plurality of structural elements, a structural element type selected from a heading type, a section type, a table type, and a reference type and a corresponding character offset range or bounding box in the input unstructured electronic document.
The computer-implemented method for converting an unstructured electronic document into an Agent Document Format container file comprising further comprising generating conversion benchmark data that quantifies at least one of an extraction duration, a structural detection duration, a semantic enrichment duration, a vector embedding generation duration, and a token usage for the large language model, and packaging the conversion benchmark data into the single compressed archive file with the semantic enrichment data and the vector embedding for each semantic chunk.