Composing Smart Data Services in Shop Floors Through Large Language Models

Recent years have witnessed an ever-growing use of Large Language Models (LLMs) to lower the technical barrier for several tasks, ranging from coding to querying relational databases to composing services. In this work, we focus on using LLMs to simplify access to data in the industrial scenario, by allowing humans operating on the shop floor to submit a query in natural language and then materializing a table integrating data gathered from different data sources including machines and information systems. In particular, we introduce COSMADS, which takes as input a query from an operator on the shop floor and automatically synthesizes a pipeline that leverages existing data sources accessible as services (data services), to compose a table output fulfilling the user’s information need. The proposed solution is evaluated using a real case study, showing that results obtained by taking into account available data service descriptions and previous pipelines outperform those obtained by naively employing a state-of-the-art code generation tool.