MUMBAI, India, June 26 -- Intellectual Property India has published a patent application (202641052356 A) filed by The Principal, Sns College Of Technology on April 24, 2026, for System And Method For Adaptive Two Stage Retrieval Augmented Generation In Automated Incident Response.
Inventors include Dr Muthu Vijaya Pandian, S; Alna, Pm; Jeegath Prakash. M; Karthick Raja, E; Kavin, R; Prajan, G; Padmapriyan. E; Srinithi, R; and Shreya. Sd.
The application for the patent was published on June 19, 2026, under issue no. 25/2026.
Abstract: ABSTRACT: An Ai-Powered Adaptive Learning Platform is disclosed for managing personalized student learning in distributed educational environments while maintaining full institutional data sovereignty. The invention utilizes a high-efficiency ingestion layer (104) featuring a Filter and Dedupe pipeline (104b) that achieves substantial interaction data compression and enables the processing of high-velocity learner activity streams from large concurrent student cohorts. A dual-path tokenizer (106a) is employed to compress learner interaction records into Skill Template IDs (106d) while preserving critical metadata, such as Learner IDs (106ca) and Subject IDs (106cb), to prevent context fragmentation across multi - subject learning journeys. The system implements an adaptive two-stage Retrieval- Augmented Generation (RAG) mechanism that dynamically switches between internal curriculum knowledge bases (110a) and an agentic content retrieval fallback (110b). This transition is governed by a configurable Cosine Similarity threshold (a = 0.7) to address knowledge gap scenarios where internal documentation is insufficient to generate a relevant personalized learning response. To maintain absolute student privacy, a Mandatory Sanitization Layer utilizing a localized, 4-bit quantized Small Language Model (SLM) (112a) performs a sensitivity sweep to mask personally identifiable information (PH) before any external content retrieval queries are initiated. The invention provides personalized study paths, concept-level micro-feedback, and practice questions to individual students through a real-time Student Dashboard (114a), while simultaneously delivering cohort-level misconception analytics, at-risk learner identification, and intervention recommendations to educators through a Teacher Analytics Dashboard (114b). The invention provides a scalable, cost-effective alternative to commercial LLM APIs by eliminating external token costs and student data leakage risks through localized computation.
Disclaimer: Curated by HT Syndication.
