Polars DataFrame AI Skill Guide

Overview & Engine Architecture

Polars is a Rust-backed DataFrame library with eager DataFrame and deferred LazyFrame plans. Expressions (pl.col, pl.when) describe column work; the optimizer pushes projections and predicates into scans. Agents default to lazy + collect() (or streaming) for multi-file Parquet pipelines and keep joins/filters as expressions, not Python loops.

scan_parquet / scan_csv
      -> LazyFrame plan
          -> filter / select / group_by / join
          -> collect() or sink_parquet()

When to use this skill

• Faster local ETL than pandas on medium/large files
• Columnar pipelines that stay on Parquet
• Expression-heavy transforms before export to DuckDB/Spark