MUMBAI, India, Aug. 10 -- Intellectual Property India has published a patent application (202641090990 A) filed by Vellore Institute Of Technology on July 27, 2026, for Retrieval-Augmented-Generation Assisted Explainable Machine Learning System For Intelligent Preprocessing And Decision Auditing.
Inventors include Dr M Thenmozhi; and Amrit Kumar Gorai.
The application for the patent was published on July 31, 2026, under issue no. 31/2026.
Abstract: ABSTRACT RETRIEVAL-AUGMENTED-GENERATION ASSISTED EXPLAINABLE MACHINE LEARNING SYSTEM FOR INTELLIGENT PREPROCESSING AND DECISION AUDITING A retrieval-augmented-generation assisted explainable machine learning system (100) for intelligent preprocessing, model recommendation, and decision auditing is provided. The system (100) comprises a data analyzer module (102) configured to receive a dataset and compute statistical characteristics of numeric columns, wherein the statistical characteristics comprise missing values percentage, skewness, and outliers percentage. The system (100) further comprises a human-in-the-loop preprocessing interface (104) configured to present the statistical characteristics to a user and receive user-selected preprocessing decisions for one or more columns of the dataset. The system (100) additionally comprises a logging module (106) configured to record analyzer insights and preprocessor insights in a structured data format, and a preprocessing auditing engine (108) configured to compare analyzer-suggested preprocessing techniques with user-selected preprocessing decisions and identify preprocessing mismatches. The system (100) further comprises a retrieval-augmented-generation explainability engine (110) comprising an embedding model (202) configured to convert the recorded insights into embedding vectors, a vector database (204) configured to store the embedding vectors, a semantic retrieval engine (206) configured to query the vector database (204), and a large language model (208) configured to generate human- readable explanations for preprocessing decisions based on retrieved embedding vectors.
Disclaimer: Curated by HT Syndication.
