MUMBAI, India, Sept. 28 -- Intellectual Property India has published a patent application (202641111571 A) filed by Subba Nelakudhiti; Rajasekhar Reddy Arikatla; Sankethu Babu Guntaka; Sneha Valusa; and Rajesh Thonduru on September 17, 2026, for A Semantic Runtime Optimization System For Energy-Efficient Large Language Model Deployment At The Edge.
Inventors include Subba Nelakudhiti; Rajasekhar Reddy Arikatla; Sankethu Babu Guntaka; Sneha Valusa; and Rajesh Thonduru.
The application for the patent was published on September 25, 2026, under issue no. 39/2026.
Abstract: The present invention provides the system incorporates an input semantic analysis engine that assesses incoming natural language queries to extract metrics related to semantic complexity, classify intents, and recognize contextual dependencies at the token level. This approach enables the avoidance of unnecessary computational layers when dealing with low-complexity inputs. Additionally, a dynamic runtime profiling module connects with hardware performance counters to track real-time telemetry data, which includes thermal conditions, power consumption patterns, and memory bandwidth across edge computing nodes. Moreover, an LLVM-based compilation framework for whole programs employs compiler wrappers to produce and integrate a unified Intermediate Representation (IR) of the LLM inference engine. To ensure optimal low-power performance while maintaining semantic integrity, a combinatorial genetic algorithm optimization engine analyzes specific code transformation processes using hardware energy benchmarking metrics as fitness values. FIG.1
Disclaimer: Curated by HT Syndication.
