The hosting looks cool but not efficient. Would you consider support batch inference for a more scalable solution?

Given that you already rely on Polars, did you consider providing a pure Rust interface to Metaflow?