Intellectual Property India Publishes Patent Application for Varman Ai: A Quantum-Enhanced Multi-Agent System Anonymization Of Personal Identifiable Information.

MUMBAI, India, Aug. 12 -- Intellectual Property India has published a patent application (202621061741 A) filed by Furiya Priyanshi Alpesh; and Nair Abhishek on May 15, 2026, for Varman Ai: A Quantum-Enhanced Multi-Agent System Anonymization Of Personal Identifiable Information..

Inventors include Priyanshi Alpesh Furiya; Abhishek Anil Nair; Dr. Ami Munshi; Dr. Vaishali Kulkarni; and Kapil Rathor.

The application for the patent was published on August 07, 2026, under issue no. 32/2026.

Abstract: A quantum-enhanced multi-agent system and method for privacy-preserving document processing, referred to as Varman AI, is disclosed. The system orchestrates a nine-component sequential pipeline enabling secure interaction between sensitive user documents and external large language models (LLMs) without exposing Personally Identifiable Information (PII). A Multi-Modal Ingestion Module processes heterogeneous inputs including PDFs, audio, images, and structured data. A Domain-Specific Routing Agent classifies documents into regulatory-aware domains using advanced prompting techniques including Chain-of-Thought and Tree-of- Thoughts. A PII Redaction Agent extracts PII using a hybridized rule-based and machine learning NER approach. A Quantum Parameterized Feature Extraction Layer employs a quantum-classical hybrid variational circuit to generate semantically coherent, constraint-satisfying synthetic entities, validated via Quantum Kernel Similarity Search to produce a numerical Privacy Score. A Relational Context and Dependency Graph preserves inter-entity structural and linguistic relationships post- anonymization. An Entity Masker replaces all PII with synthetic counterparts. An Adversarial Critic Agent validates masked output using deterministic evaluation and comparison metrics including BERT-score and phonetic hashing, triggering iterative re-masking until zero-leakage is verified. The validated masked text is forwarded to an external LLM and the response is de-anonymized by reverse-substitution. The system incorporates an adaptive self-learning feedback loop using RLHF and DPO, HMAC-based session authentication, persistent session storage, and multi-environment deployment support.

Disclaimer: Curated by HT Syndication.