MUMBAI, India, Nov. 2 -- Intellectual Property India has published a patent application (202611109025 A) filed by Graphic Era Deemed To Be University on September 11, 2026, for A System And Method For Contextual Reconstruction And Explainable Retrieval-Augmented Long- Document Understanding.
Inventors include Arun Singh; Vartika Agarwal; Gagan Bansal; and Geetika Sharma.
The application for the patent was published on October 30, 2026, under issue no. 44/2026.
Abstract: The invention relates to computer-implemented long-document analysis and retrieval-augmented generation for PDF documents. An offline ingestion pipeline extracts page text, omits low-content pages, forms adaptive section node records linked in document order, and decomposes pages into sequential atoms of about 50 to 100 words or tokens, each atom being cross-referenced to a parent section and serialized with the section records under a stable document identifier. A real time query pipeline prompts a language model to select one to three relevant section nodes, restricts vector similarity search to atoms within the selected sections, seeds anchor atoms, and reconstructs context through multi-pass neighboring atom expansion, deduplication, sequential ordering, gap marking, and concatenation into a reconstructed narrative block. The system advances existing retrieval methods by reducing search space, preserving document continuity, and generating explainable provenance including macro reasoning paths and micro evidence chains for answer generation, audit logging, and authorized multi-tenant document access.
Disclaimer: Curated by HT Syndication.
