Intellectual Property India Publishes Patent Application for A Transformer-Orchestrated Multimodule System For Interventional Reward Estimation And Constraint-Aware Real-Time Content Retrieval

MUMBAI, India, Aug. 24 -- Intellectual Property India has published a patent application (202641098955 A) filed by Dayananda Sagar University on August 16, 2026, for A Transformer-Orchestrated Multimodule System For Interventional Reward Estimation And Constraint-Aware Real-Time Content Retrieval.

Inventors include Ankita Himmatlal Thakkar; M. Lakshmanan; Trupthi Rao; Jayavrinda V. Vadakkeparambil; Manikandan Moovendran; Sriramkumar R; Mithaguru; Prateek Verma; Govind Kumar Pandey; Srinidhi Kuna; and Shivnandan Rai.

The application for the patent was published on August 21, 2026, under issue no. 34/2026.

Abstract: Title: A Transformer-Orchestrated Multimodule System for Interventional Reward Estimation and Constraint-Aware Real-Time Content Retrieval The present invention relates to a processor-implemented transformer-orchestrated multimodule system for real-time content retrieval and adaptive recommendation generation. The system comprises a Feature Fusion Module (102), a Causal Treatment-Effect Estimation Module (103), a Transformer-Based Orchestration Engine (104), a Contextual Multi-Armed Bandit Module (105), a Fairness and Constraint Enforcement Module (106), a Feature Attribution Module (107) and a Closed-Loop Policy Update Module (110). The transformer applies attention to module-level information to generate a bounded Adaptive Multi-Policy Vector that coordinates exploration– exploitation, causal decision-making, fairness enforcement, explainability and policy adaptation. Recommendation candidates are selected using causal treatment-effect estimates and contextual optimization, validated against predefined constraints and accompanied by feature-level explanations. Runtime feedback continuously updates policy parameters without repeated full-model retraining, thereby improving adaptability to changing user behaviour and non-stationary data distributions while reducing computational overhead. The invention provides a scalable, computationally efficient, fairness-aware, explainable and causally informed framework for real- time content retrieval and recommendation systems.

Disclaimer: Curated by HT Syndication.