MUMBAI, India, July 7 -- Intellectual Property India has published a patent application (202421104414 A) filed by Lti Mindtree Ltd. on December 30, 2024, for Method And System For Improving Retrieval Accuracy In Retrieval Augmented Generation (rag) Framework.
Inventors include Dr. Anindita Desarkar; Aritra Sen; and Vishwanathan Raman.
The application for the patent was published on July 03, 2026, under issue no. 27/2026.
Abstract: Methods and systems for improving retrieval accuracy in a Retrieval Augmented Generation (RAG) framework are disclosed. Input documents (102) are analyzed using a Small Language Model (SLM) to identify document types and extract content using type- specific methods. An adaptive chunking process generates chunked content based on variables including use case type, speed, cost, latency requirements, and Large Language Model (LLM) context window size. The chunked content undergoes tokenization using strategies based on accuracy and vocabulary requirements. An embedding model is selected based on user requirements, and quantization policies are applied to generate optimized vector representations. Document portions are retrieved using a multi-step strategy comprising hybrid search, meta filters, multi-query retrieval, and hypothetical document embeddings. Performance is evaluated through context relevance, groundedness, and answer relevance scores, implementing parameter adjustments and re- evaluation processes when scores fall below thresholds.
Disclaimer: Curated by HT Syndication.
