I have written a description of a "pandas-like" API for mlrun which would provide a consistent and modular API which is easy to use interactively. The first description can be found here. There is some additional implementation for nuclio runtime autocomplete support here - this has implementation / docstrings if you're interested. The main idea is to reduce documentation and learning curve through highly expressive, consistent code. I have started with an idea for the API, but I'm happy to contribute to the refactor (implementation and testing). Please let me know your thoughts @Hedingber and @yaronha!