Build Your Own DataFrame: a course based on an engine I probably shouldn't have written

A few years ago, I needed a data processing engine for a visual ETL tool I was building — Flowfile — and against all sane practices, I just started writing one. Pure Python. itertools.groupby for aggregation. operator.itemgetter for column access, own type inference, manual memory optimization, custom everything. It handled joins, pivots, groupby, explode, filters — a working engine built entirely on the standard library. No numpy, no C extensions, no dependencies at all. Was this a good idea? Probably not the most efficient path. But it taught me something I couldn't have learned any other way: I understood exactly what a dataframe library does, because I'd built every piece of one myself. When I eventually migrated Flowfile's engine to Polars, the pure Python engine went into a drawer. That migration was driven by something I realized about focus: you can't do everything. Building a custom dataframe engine was a great way to learn, but it was a terrible way to ship a product. Flowfil