MUMBAI, India, June 30 -- Intellectual Property India has published a patent application (202641074417 A) filed by Rajarshi Ghosh Dastidar on June 16, 2026, for Ai-Driven Agentic System For Automated Generation, Validation, Repair, And Orchestration Of Executable Dag Workflows.
Inventor includes Rajarshi Ghosh Dastidar.
The application for the patent was published on June 26, 2026, under issue no. 26/2026.
Abstract: AI-DRIVEN AGENTIC SYSTEM FOR AUTOMATED GENERATION, VALIDATION, REPAIR, AND ORCHESTRATION OF EXECUTABLE DAG WORKFLOWS 5 10 15 20 25 An AI-driven agentic system for automated generation, validation, repair, and orchestration of executable directed acyclic graph (DAG) workflows, including quantitative trading strategy workflows. The system comprises cooperating subsystems: (1) a catalog-injected natural language DAG generation subsystem that dynamically retrieves the platform's live node catalog, embeds quantitative finance domain knowledge and guardrails into a large language model prompt, and produces structurally valid trading strategy graphs from natural language instructions; (2) a dual-format DAG bridging subsystem that bidirectionally transforms between a visual canvas representation and a C++ computational engine manifest format with structural validation at each boundary; (3) a state-machine workflow orchestration subsystem that inspects workspace state and computes context-sensitive next-action recommendations for guided strategy development lifecycle progression; and (4) an artifact provenance and truth-status subsystem that maintains producing engine identifiers, lifecycle stage classifications, data source lineage, chronological action chains, and real/synthetic/blocked evidence status for every artifact; (5) a bounded validation-and-repair subsystem; (6) a permission-scoped tool and approval-gating subsystem; and (7) a persistent multi-page context subsystem; (8) strategyspecific DAG canonicalization; (9) topology repair driven by executable backend feedback; and (10) lifecycle-aware agent routing across build, backtest, sweep, machine-learning research, multi-stage research, and deployment modes. A non-transitory computer-readable medium storing instructions for performing the disclosed methods is also claimed. In certain embodiments, the system extends a graphical platform for end-to-end predictive model and trading logic construction.
Disclaimer: Curated by HT Syndication.
