MUMBAI, India, Aug. 31 -- Intellectual Property India has published a patent application (202621083273 A) filed by Ronakkumar Shah on July 07, 2026, for A Hybrid Semi-Automated Workflow For Systematic Literature Reviews Using Large Language Models..
Inventor includes Ronakkumar Shah.
The application for the patent was published on August 28, 2026, under issue no. 35/2026.
Abstract: ABSTRACT [501] Systematic literature reviews are foundational to evidence-based research across medicine, public health, software engineering, and social science, yet the conventional review process is severely constrained by the volume of literature requiring screening, the time demanded of expert reviewers, and the inherent risk of inconsistent application of eligibility criteria across thousands of records. Existing review-support tools either provide passive reference management with no intelligent screening capability, or deploy fully automated machine-learning classifiers that operate without a structured human verification checkpoint, exposing the review to undetected systematic errors and limiting their acceptance for publication-grade evidence synthesis. [505] The present invention discloses a Hybrid Semi-Automated Workflow for Systematic Literature Reviews using Large Language Models, comprising a modular architecture integrating a Literature Retrieval Module for multi-database search execution and deduplication, an LLM Screening Module that applies user-defined eligibility criteria to generate inclusion/exclusion decisions accompanied by machine-generated justification text, a Human-in-the-Loop Verification Module enabling rapid reviewer confirmation, override, or escalation of every automated decision, a Data Extraction Module that performs LLM-assisted structured field population with source- text traceability, and an Audit Trail and Reporting Module that continuously logs all automated and human decisions and auto- generates a PRISMA-compliant flow diagram. [510] The system distributes the screening and extraction burden between the Large Language Model and the human reviewer such that the reviewer's task is reduced from independently classifying every record to verifying a machine-generated recommendation, while retaining a documented human decision at every stage required for scientific rigor. Validation through seven-stage sequential development including requirement analysis, modular system design, retrieval integration, screening-pipeline development, verification-interface development, extraction and audit integration, and calibrated testing against gold-standard reviews confirms that the invention constitutes a novel, transparent, and reviewer-accountable approach to accelerating systematic literature review production without compromising methodological integrity.
Disclaimer: Curated by HT Syndication.
