Bai, Jiadong; Huang, Yicong; Li, Chen
VLDB (Demo track), 2026.
Recent advances in large language models (LLMs) have made human-agent collaboration a promising and powerful paradigm for data science. Among existing LLM-based solutions, conversational analytics is convenient but makes it difficult for users to inspect intermediate steps and verify errors. Script-based ReAct improves transparency by grounding interaction in coding scripts, but still inherits the complexity and low-level details of programming code. In this paper, we demonstrate a novel ReAct-based system called BobFlow, which is built on the idea of using dataflow as an abstraction for human-agent collaboration. This abstraction is especially friendly for LLMs to do their internal reasoning for data science tasks. In the demo, we will show the benefits of BobFlow developed on top of Apache Texera (Incubating), including how an analyst easily understands the agent’s behaviors through an intuitive dataflow-based interface, how the analyst efficiently inspects the agent’s past actions to provide feedback, and how the agent finishes a task with high accuracy and low cost.
