MUMBAI, India, Jan. 23 -- Intellectual Property India has published a patent application (202521105474 A) filed by Persistent Systems, Pune, Maharashtra, on Oct. 31, 2025, for 'system and method for query-conditioned feature extraction and estimation-based attention across large-scale repositories.'
Inventor(s) include Mr. Nitish Shrivastava; Mr. Pradeep Kumar Sharma; and Mr. Shantanu Godbole.
The application for the patent was published on Jan. 23, under issue no. 04/2026.
According to the abstract released by the Intellectual Property India: "The present invention relates to a system and method for computing query-conditioned attention across a large set of files in complex repositories without the use of large language models (LLMs). The invention employs multi-dimensional feature extraction and projection techniques to estimate relationships between repository files and a user query. By constructing an attention matrix based on engineered feature vectors rather than deep neural token embeddings, the invention provides a scalable, interpretable, and computationally efficient mechanism for identifying the most relevant files or components in response to a query. The invention further introduces a sparse attention mechanism that dynamically selects attention neighborhood using lexical, structural, and behavioral repository data, enabling high-fidelity relevance estimation on repositories exceeding tens of thousands of files."
Disclaimer: Curated by HT Syndication.
