MUMBAI, India, Sept. 14 -- Intellectual Property India has published a patent application (202621090977 A) filed by Vaishali Dayanand Suryawanshi; Dr. Dayanand Suryawanshi; Ragita Nai; Atish Ashokrao Shriniwar; Vinayak Sambhaji More; Sumit Subhash Sasane; Dr. Murthy Ravaleedhar Reddy; Vrushabh Rajeshwar Bawankar; Dnyaneshwar Seetaram Thorat; Dr. N. Subathra; Dr. Arun Kumar Sao; and Ms. Sowmiya L on July 27, 2026, for A Context-Aware Language Model Adaptation System Using Retrieval-Augmented Generation And Dynamic Fine-Tuning.
Inventors include Vaishali Dayanand Suryawanshi; Dr. Dayanand Suryawanshi; Ragita Nai; Atish Ashokrao Shriniwar; Vinayak Sambhaji More; Sumit Subhash Sasane; Dr. Murthy Ravaleedhar Reddy; Vrushabh Rajeshwar Bawankar; Dnyaneshwar Seetaram Thorat; Dr. N. Subathra; Dr. Arun Kumar Sao; and Ms. Sowmiya L.
The application for the patent was published on September 11, 2026, under issue no. 37/2026.
Abstract: The present invention relates to A Context-Aware Language Model Adaptation System Using Retrieval-Augmented Generation and Dynamic Fine-Tuning for generating accurate, relevant, updated, and domain-specific responses. The system is designed to overcome the limitations of conventional language models that depend mainly on pre-trained knowledge and may produce outdated, generic, or unsupported information. The proposed system comprises a user input interface, context analysis module, query understanding module, knowledge retrieval module, knowledge repository, Retrieval-Augmented Generation module, dynamic fine-tuning module, response validation module, context memory module, feedback processing module, and final response delivery unit. In operation, a user query is received and analyzed to identify user intent, domain, conversation history, response style, technical depth, and access requirements. The retrieval module searches relevant information from structured and unstructured knowledge sources such as documents, databases, research papers, manuals, policy files, enterprise repositories, and web- based sources. The retrieved information is ranked according to relevance, reliability, freshness, and semantic similarity. The Retrieval-Augmented Generation module then combines the user query, contextual information, and retrieved knowledge to generate a grounded response. The dynamic fine-tuning module continuously adapts the language model using domain-specific data, updated knowledge, interaction patterns, and user feedback. The response validation module checks the generated output for factual support, completeness, consistency, and relevance before delivery. The invention reduces hallucination, improves factual accuracy, supports personalized communication, and enables real-time adaptation. The system is suitable for educational, healthcare, legal, enterprise, customer support, research, and intelligent automation applications.
Disclaimer: Curated by HT Syndication.
