Retrieval-Augmented Generation for Autonomous SOP Creation: A Methodology for Document Synthesis Across Business Functions

This Digital Operations for various business functions like Finance & Accounting, Customer Care, Supply Chain Management and others, performed by human operators are extremely reliant on sequential process steps captured in Standard Operating Procedures (SOPs). However, the creation of SOPs is highly cumbersome, resource-intensive (cost, time & effort), and time consuming. The motivation of this research is to assist human operators with an autonomous approach for creating baseline SOPs by leveraging existing SOPs in enterprise knowledge corpus. This experimental research aimed at developing a generalized methodology that can ingest SOPs in different formats and generate baseline SOPs in pre-defined business function-specific formats. The research yielded real-world scalable business applications of Retrieval Augmented Generations with Integrated Toolset design pattern at optimal cost. For the production-grade operationalization of such applications, the study focused on a structured approach involved in the selection of search algorithm during the retrieval process. The study mentions the performance of different search algorithms like Hierarchical Navigable Small Worlds (HNSW), Exhaustive K-Nearest Neighbors (X-KNN), and Full-Text search based on evaluation parameters like Latency, Cost and Accuracy. These experimental analyses unfolded the document synthesis system's capability to leverage latent process steps hidden in existing SOPs and reduce the cycle time, operational cost and effort involved in creating SOPs.