MUMBAI, India, Aug. 31 -- Intellectual Property India has published a patent application (202611072438 A) filed by Noida International University on June 11, 2026, for System And Method For Llm-Based Virtual Voter Simulation With Adaptive Persona Modelling And Bias Mitigation.

Inventors include Birendra Kumar; Vimal Bibhu; Chetan Kumar Sharma; Sujata Kumari; and Rajeev Kaushik.

The application for the patent was published on August 28, 2026, under issue no. 35/2026.

Abstract: A system and method for LLM-based virtual voter simulation with adaptive persona modelling and bias mitigation. The system comprises a hybrid persona generation module using ensemble LLM architecture with multiple models, stratified sampling with diversity preservation, and synthetic augmentation. An adaptive behavioural modelling engine implements multi-step reasoning framework integrating information acquisition, ideological reasoning, social influence, and candidate evaluation, with dynamic persona evolution based on simulated experiences. An integrated bias detection and mitigation framework continuously monitors partisan, demographic, geographic, and cultural biases, applying adversarial debiasing, calibration algorithms, and variance preservation techniques. A simulation environment models social network dynamics, influence propagation, and temporal evolution. A multi-dimensional validation system ensures accuracy across demographics, geography, and time. Hardware implementation includes multi-GPU processing clusters, distributed memory architecture, and specialized accelerators providing technical effects. The system achieves superior diversity preservation (Macro-F1 within 5% of Micro-F1), reduced partisan bias (60-75% reduction), enhanced response heterogeneity (variance within 10% of real data), and high prediction accuracy ( 2% error for national outcomes), addressing all major limitations of prior art while ensuring compliance with Indian Patent Act Section 3(k) through hardware- implemented architecture delivering technical effects beyond algorithm execution.

Disclaimer: Curated by HT Syndication.