The Tabular Data Reasoner is a tool capable of reasoning about tabular data using a pre-trained language model. It can be used to answer questions in natural language regarding data in tables and orchestrate data preparation pipelines.

Based on the provided tabular data, the tool will devise a plan composed of multiple steps. Each step in the plan is carried out by an atomic operation from a pre-defined set of data processing operators. We devise supervision and encoding techniques to effectively process tabular data and to select the appropriate operations for the corresponding question and table pairs.

A web-based API for processing input documents and questions and outputting is planned.

This service is currently under development.