DataFlow is an open-source system for easy data preparation with LLM-based operators and pipelines. Instead of hand-writing cleaning and transformation code, you compose operators that use language models to prepare training and evaluation data, with companion projects for multimodal data, reusable skills and a web UI.
The project comes out of the OpenDCAI group with a SIGMOD'27 paper behind it, and everything lives in the open on GitHub - free to use, extend and self-host.
