Leveraging LLMs for Context-Aware Process Mining and Analysis on Multi-Source Data

Organizations are fundamentally based on business processes that allow the streamlined functioning and achievement of goals. These business processes depend on various IT systems and tools that manage their performance. While techniques like process mining enable the analysis of these processes, the presence of multiple sources of contextually linked data makes it difficult to derive valuable insight with ease. This research aimed to establish contextual connections between multiple sources of data and perform information retrieval and analytical tasks using LLMs with a focus on process mining. We introduce a novel approach that utilizes a custom-built Python-based Retrieval Augmented Generation (RAG) pipeline that has been designed to overcome the limitations of existing LLM-based approaches, mainly focusing on handling multiple data sources and process-oriented queries. We have incorporated a unique table and row retrieval mechanism, providing the LLM with rich context and enabling it to reason about data from a process-aware perspective. In conclusion, this paper proves the ability to contextually link process logs with additional process-related data from varied sources to enable LLMs in providing relevant process insights which can be used to plan, monitor, and improve operations all through natural language querying.