SynergyAI: A Human–AI Pair Programming Tool Based on Dataflow

Abstract

This paper proposes SynergyAI, an AI–human pair programming tool that represents predictive models as dataflows composed of AI, input, and output nodes. By visualizing decision tree models and integrating them with dataflow diagrams, SynergyAI effectively addresses the machine learning black-box problem. Additionally, the tool leverages comprehensive prediction algorithms and ensemble learning to simplify the operation of complex dataflows and mitigate overfitting risks. SynergyAI also features an AI assistant that utilizes scatter plot matrices and data correlation analysis to help programmers select data and optimize model structures. Experimental results demonstrate that, through human–AI collaboration, SynergyAI achieved an accuracy of 85% in predicting mechanical failures in a chocolate factory, providing an efficient and powerful tool for programming and data analysis.

Keywords

• data flow
• decision tree
• human–AI collaboration
• explainability