MUMBAI, India, July 24 -- Intellectual Property India has published a patent application (202621065504 A) filed by Nilanjan Shankar Paul; and Janhavi Dahatonde on May 20, 2026, for A Lifecycle-Aware Hybrid Cognitive Memory Architecture For Persistent Neural Retrieval And Multi-Hop Reasoning.
Inventors include Janhavi Dahatonde; and Nilanjan Shankar Paul.
The application for the patent was published on July 17, 2026, under issue no. 29/2026.
Abstract: The lifecycle-aware hybrid cognitive memory architecture described in this invention is designed to support persistent artificial intelligence systems and long-term neural retrieval applications. The framework includes three types of memory: working, episodic, and semantic, for continuous contextual reasoning, multi-session continuity of interaction, and adaptive retrieval of information in large language model (LLM)-based systems. This memory architecture combines two types of retrieval: similarity based on dense vectors and relational reasoning using graphs, to allow for both semantic understanding and multi-hop inferencing in long-context interactions. The hybrid memory architecture uses multi-vector embedding representations of content, entity, relationship, and context to improve retrieval accuracy and to create contextual associations between items in memory. This invention provides a mechanism for managing the lifecycle of memory using adaptive techniques such as temporal decay, reinforcement learning, pruning, and consolidation through abstraction in order to maximize retention while minimizing redundancy and increasing computational efficiency over long periods of time. Additionally, a new mechanism for determining the composite score of retrieved items based on their semantic relevance, recency (in terms of time), frequency (of retrieval), and strength (of relational connection) will allow for dynamic rankings of memory candidates. Through the use of the invention, ongoing context awareness and the ability to reuse previously processed data can be achieved, along with improved long-term memory management for artificial intelligence systems and enhanced reasoning challenges over an extended period of time. Both conversational AI agents and intelligent assistant applications will benefit from this proposed architecture as will adaptive tutoring systems, enterprise knowledge systems, autonomous agents, health care AIs, and cyber security systems; therefore, creating another long-context reasoning platform.
Disclaimer: Curated by HT Syndication.
