MUMBAI, India, Sept. 14 -- Intellectual Property India has published a patent application (202611086563 A) filed by Noida Institute Of Engineering And Technology Niet on July 15, 2026, for System And Method For Privacy-Preserving Adaptive Prompt Compression Using Edge-Deployed Retrieval-Augmented Graph Analysis.
Inventors include Neeti Taneja; and Dr. Ritesh Kumar Singh.
The application for the patent was published on September 11, 2026, under issue no. 37/2026.
Abstract: A privacy-preserving adaptive prompt compression system (100) deployed on an edge computing device compresses input prompts for large language model inference. The system (100) comprises an edge RAG compressor module (110) leveraging locally stored retrieval indices for composite token scoring combining cosine similarity with inverse redundancy, a privacy sanitization engine (120) applying entity-sensitivity-calibrated differential privacy noise on embedding representations and token obfuscation with local de-obfuscation mapping, an adaptive compression scheduler (130) with neural network-based workload-aware compression ratio prediction, a multi-layer dependency analyzer (140), a compressed prompt assembler (150), and a reinforcement learning-based feedback loop controller (160). All processing occurs at the edge device, producing technical effects of reduced data transmission, reduced latency, and prevented sensitive data exposure. [Figure 1]
Disclaimer: Curated by HT Syndication.
