MUMBAI, India, Aug. 12 -- Intellectual Property India has published a patent application (202621074491 A) filed by Dheeraj Kumar Singh on June 16, 2026, for Privacy-Preserving Local Gpu/Cpu-Based Retrieval Augmented Generation System.
Inventor includes Dheeraj Kumar Singh.
The application for the patent was published on August 07, 2026, under issue no. 32/2026.
Abstract: The present invention relates to a privacy-preserving local Retrieval-Augmented Generation apparatus using zero-retention, session-isolated processing. The apparatus includes a processing unit, volatile memory, user interface, session manager, document processing module, semantic chunking engine, embedding generation module, in-memory vector index, Top-K retrieval engine, controlled inference layer, local language model engine, adaptive CPU–GPU controller, and memory sanitization module. A user document is loaded into volatile memory, processed locally, divided into semantic chunks, converted into embeddings by a locally deployed embedding model, and stored only in a session-bound in-memory vector index. A user query is encoded locally, and only Top-K relevant chunks are retrieved for constructing a context-limited prompt. The local language model generates a grounded response from the retrieved content without cloud transmission. Upon session completion or timeout, the apparatus invalidates the session identifier and erases documents, embeddings, vector indexes, queries, prompts, and buffers, thereby preventing persistent data retention and cross-session contamination. (Accompanied Figure No. 1-7)
Disclaimer: Curated by HT Syndication.
