Manufacturing enterprises, especially those in discrete and repetitive sectors, continually generate vast and heterogeneous data streams, such as machine logs, ERP/MES records, and supply-chain documents, that frequently remain siloed and fragmented. Integrating these disparate data sources to build unified pipelines is critical for enabling process optimization and minimizing operational losses. Traditional integration efforts often require third-party services, incurring significant costs and delays with extreme manual efforts. However, recent advances in artificial intelligence (AI), particularly large language models (LLMs), signal a transformative shift in how manufacturing organizations can overcome data silos.

State-of-the-art LLMs now support automatic schema suggestion, fallback rules, and labor-saving ETL workflows, dramatically reducing manual integration efforts.AI-assisted platforms increasingly offer connectors and schema-less extractors for ingesting data from diverse sources, including PDFs, images, and IoT logs, into LLM-powered agentic pipelines. The presentation demonstrates one such use-case in injection molding process to extract and merge contextual shopfloor and enterprise information in real time, thereby supporting responsive scheduling, predictive maintenance, and tighter quality control. Importantly, layering AI on legacy PLC/ERP infrastructure enables manufacturers to extract new insights while preserving existing investments.