MUMBAI, India, Aug. 10 -- Intellectual Property India has published a patent application (202621070833 A) filed by Nehal Shah; and Priti Nehal Shah on June 07, 2026, for Hybrid Large Language Model (llm) And Deterministic Policy Engine For Explainable Real-Time Tax Classification.
Inventors include Nehal Shah; and Priti Nehal Shah.
The application for the patent was published on July 31, 2026, under issue no. 31/2026.
Abstract: A hybrid large language model and deterministic policy engine for explainable real-time tax classification is disclosed. The system combines a fine-tuned large language model for probabilistic classification of ambiguous financial events with a deterministic, versioned policy engine for statutory tax computation. A contextual feature vector augmenting each financial event with historical transaction patterns enables accurate classification. A confidence-gated verification trigger suspends computation and generates citizen verification requests for low-confidence classifications. A routing engine maps classification outputs to applicable tax computation rules. A dual-source explainability generator combines template-based statutory explanations with LLM-generated plain-language explanations. A privacy-preserving feedback learning loop improves classification accuracy using anonymized citizen acceptance/rejection signals. All computation outputs are cryptographically signed and sealed in an immutable evidence vault. The hybrid architecture ensures computations are explainable, reproducible, and regulatory-defensible, while handling the full spectrum of deterministic and ambiguous tax classification scenarios.
Disclaimer: Curated by HT Syndication.
