Agentic cognitive orchestration for cross-lingual patent intelligence: A modular workflow automation approach

Background Global patent intelligence demands continuous monitoring of multilingual policy announcements from national intellectual property offices (IPOs); however, source heterogeneity, language barriers, and fragmented data structures create significant information asymmetry, particularly for non-English jurisdictions. Existing solutions remain fragmented and reliant on manual batch processing, leaving organizations exposed to critical blind spots in global competitive intelligence. Objective This study proposes and empirically evaluates an Agentic Cognitive Orchestration (ACO) framework that automates cross-lingual patent intelligence monitoring by integrating low-code workflow automation with large language models (LLMs) into a reproducible, modular pipeline. Methods The ACO framework orchestrates five major patent offices—EPO, USPTO, WIPO, TIPO, and MOIP—through a five-stage sequential pipeline: (1) RSS Feed Aggregation and Data Ingestion via a Dual-Path strategy combining protocol-based acquisition for structured sources and deep-scraping for unstructured HTML sources; (2) Date Normalization with Unified Timestamp Formats; (3) an LLM-based AI Agent for Language Detection and Translation; (4) Semantic Metadata Generation and Content Summarization; and (5) Multi-format Newsletter Generation and Intelligent HTML Templating. A Cognitive Triage layer employing chain-of-thought (CoT) prompting with entity-locking suppresses hallucination risk for critical patent entities such as patent numbers, dates, and legal citations. System performance was evaluated on 340 announcements collected from five jurisdictions over a four-month period. Results The ACO framework effectively filtered administrative noise, while preserving high-value strategic intelligence across English, Korean, and multilingual sources. Automated processing reduced multilingual content aggregation from 4 to 6 h of manual effort to near real-time delivery. The pipeline adheres to FAIR reproducibility principles and is fully encapsulated as a portable JSON artifact for replication. Qualitative evaluation confirmed the system's capacity to extract culturally specific IP narratives and structured impact assessments from non-English jurisdictions. Conclusions Integrating workflow automation with constrained LLMs provides a scalable, scientifically reproducible approach to real-time cross-lingual patent intelligence. The proposed framework democratizes global IP surveillance capabilities previously accessible only to large organizations with dedicated intelligence teams, offering direct implications for R&D strategy, competitive intelligence, and IP portfolio management in small and medium-sized enterprises.