A Unified Agentic and Generative AI Framework for Enhancing Drilling Intelligence Through WITSML and Unstructured Data (pdfs) Interaction

This paper presents a proof-of-concept (PoC) implementation of a Generative Al-driven architecture that enables intuitive access to both structured data, such as WITSML logs, and unstructured content, such as PDFs, through natural language interaction. Leveraging the proprietary InsightsAI platform, the solution integrates domain-specific ontologies and a knowledge graph to semantically correlate data from different sources. A data virtualization layer provides federated access across heterogeneous sources, while GenAI-powered interfaces allow domain experts to query datasets conversationally, with outputs delivered in visual or tabular formats. Contextual enrichment logic enhances real-time drilling parameters with historical operational context, and agentic AI routines autonomously identify relevant datasets and perform multi-step reasoning to synthesize insights. Validation using the publicly available Volve dataset demonstrated seamless cross-domain data discovery, efficient retrieval of complex drilling information, and improved situational awareness through semantic correlation. The modular architecture reflects real-world upstream operational environments where rapid, multidisciplinary decision-making is essential. By combining Large Language Models (LLMs), Small Language Models (SLMs), agentic AI, and machine learning with semantic data integration, this approach accelerates decisions, reduces manual effort, and enhances knowledge reuse—delivering scalable intelligence and measurable ROI in oil and gas exploration and drilling operations.