Intellectual Property India Publishes Patent Application for “a Content-Addressed Dag Execution Engine With Trigger-Filtered Evaluation, Recursive Feature Hashing, And Rolling-Window Model Scheduling For Quantitative Trading Systems”

MUMBAI, India, June 30 -- Intellectual Property India has published a patent application (202641075713 A) filed by Rajarshi Ghosh Dastidar on June 18, 2026, for “a Content-Addressed Dag Execution Engine With Trigger-Filtered Evaluation, Recursive Feature Hashing, And Rolling-Window Model Scheduling For Quantitative Trading Systems”.

Inventor includes Rajarshi Ghosh Dastidar.

The application for the patent was published on June 26, 2026, under issue no. 26/2026.

Abstract: A CONTENT-ADDRESSED DAG EXECUTION ENGINE WITH TRIGGER-FILTERED EVALUATION, RECURSIVE FEATURE HASHING, AND ROLLING- WINDOW MODEL SCHEDULING FOR QUANTITATIVE TRADING SYSTEMS The present invention relates to a content-addressed directed acyclic graph (DAG) execution engine for quantitative trading systems. The engine comprises cooperating subsystems: (1) a recursive feature hashing subsystem that computes cryptographic content addresses over the full transitive dependency tree of each node for memoized computation without explicit version management; (2) a trigger-filtered topological evaluation subsystem that uses bitmask-based trigger filtering to enable mixed-frequency node coexistence in a single graph; (3) a rolling-window model scheduling subsystem that embeds date-indexed weight sets within ML inference nodes and switches active weights through zero-copy pointer reassignment; (4) a tick- level backtesting subsystem that uses binary-search timestamp alignment for lookahead-free depth-aware fill simulation with multi-leg pairs order support and position carry across rolling fold boundaries; and (5) a polymorphic node type subsystem with tagged union outputs, semantic feed category annotations, and a dual-mode indicator architecture that automatically broadcasts scalar operations across universe vectors; (6) an execution contract and live risk-envelope subsystem; (7) a kill- switch and operational safety state machine; (8) a pair-ladder beta-hedged execution subsystem; and (9) a model-health, A/B comparison, and unified job-control subsystem; (10) lazy per-symbol universe-vector state allocation; (11) parameter-sweep execution with isolated state and refinement ranking; and (12) configurable execution reference modes for fill simulation A/B comparison. A non-transitory computer-readable medium storing instructions for performing the disclosed methods is also claimed. The engine provides a computational foundation for GUI-based and AI-assisted quantitative trading platforms.

Disclaimer: Curated by HT Syndication.