In the context of Industry 4.0, data pipelines are increasingly central to the integration of systems, decision support, and real-time industrial automation. One critical capability is understanding and tracing how data flows and transforms through ETL (Extract, Transform, Load) stages. Existing approaches provide limited semantic visibility, relying mostly on logs, static DAGs, or limited metadata tagging. This paper presents a practical and reproducible implementation of a semantic-aware data lineage tracking system, using graph modeling (Neo4j), ontology representation (RDF), and semantic querying (SPARQL). We detail the design, implementation, and experimental validation of the pipeline, using the AdventureWorks DW2022 dataset. Results show the effectiveness of the model in enhancing traceability, querying flexibility, and integration-readiness for industrial-scale deployments.
Real-time biomedical monitoring systems generate continuous streams of physiological data that must be integrated into analytical repositories with minimal latency. Traditional Extract–Transform–Load (ETL) pipelines used in healthcare infrastructures typically rely on static synchronization schedules that cannot adapt to bursty workloads or fluctuating computational resources. This often leads to increased latency, stale analytical views, or inefficient resource usage. This paper proposes a reinforcement learning-driven adaptive ETL orchestration framework designed for biomedical signal streams in intensive care environments. The proposed approach models synchronization scheduling as a sequential decision problem where an agent dynamically adjusts synchronization intervals and batch sizes based on real-time telemetry, including latency, backlog size, throughput, and system resource utilization. The architecture integrates change-data-capture (CDC) mechanisms and idempotent database operations to ensure incremental synchronization and consistency. Experimental evaluation using simulated ICU-style workloads demonstrates that the proposed approach improves data freshness and reduces synchronization latency compared with fixed-interval and heuristic baseline strategies. The results highlight the potential of reinforcement learning-based orchestration for improving the performance and scalability of biomedical data infrastructures, supporting more reliable real-time clinical monitoring and decision-support systems.