Driving Operational Excellence in Upstream Oil & Gas with Real-Time Data, AI, and Actionable Explainability

What this post added

This post introduces and details the technical architecture and implementation of Retrieval-Augmented Explainability (RAE) and the Expert Companion (EC). RAE involves embedding real-time time-series data into vector representations using foundation models, storing them in dual vector stores (general time series and expert-annotated). It then retrieves top-k similar cases, quantifies uncertainty, and re-ranks them based on annotation probability and pattern similarity. Finally, an LLM generates verbalized, context-rich summaries and decision recommendations. The EC complements RAE by capturing expert annotations and operational context from real-time events to continuously refine the system's learning and retrieval processes.