Uso de modelos generativos para apoyar la creación y optimización de flujos de trabajo de ciencia de datos

Abstract

Abstract: Generative models, such as large-scale language models (LLMs), offer new insights for streamlining the creation and optimization of data science workflows. This project aims to leverage these capabilities to generate specific recommendations and facilitate the creation and analysis of data sets. Specifically, LLMs have been used to implement three specific data science tasks: 1) synthetic dataset generation, 2) exploratory data analysis, and 3) code analysis, debugging, and explanation with AI. To interact with the language models, the APIs they provide have been used, and different prompts have been designed for each task. In each case, a feature and web interface were implemented to facilitate communication between the generative model and the user, thus supporting the creation of more agile workflows. The end result is a prototype that integrates these features to help analysts process data sets and create workflows more efficiently, potentially transforming the practice of data science.