Speaker
Description
The ECB Data Portal provides access to all official ECB statistics, which is a vast amount of data. Due to the big amount of timeseries published, less experienced users may have difficulties in retrieving the right information and understanding the meaning and relevance of the data. Traditional search methods can be inefficient, especially for users unfamiliar with statistical terminology, dataset structures, and SDMX metadata. There is a clear need for a more intuitive, user-friendly interface that can understand and respond to queries about economic indicators and time series data using natural language.
We introduce the ECB Data Bot interface that can understand complex queries, analyse statical data, metadata and data structures and provide precise time series, along with contextual information and sources.
This presentation will explain how the ECB Data Bot integrates LLMs and RAG techniques specifically for the ECB Data Portal requirements. We will discuss the technical setup of the overall system, illustrate the design of the RAG component, how the bot leverages in SDMX metadata, explain the evaluation methodology and present our insights and findings.