Analyzing Data

This is a data warehouse query assistant that caches what it learns. Instead of rediscovering table schemas and query patterns every time, it maintains concept mappings (like "customer" maps to CUSTOMERS table, key column user_id) and pattern libraries for common question types. The workflow is pattern lookup first, then concept lookup, then table discovery if needed, always followed by caching learnings before presenting results. It runs queries through a persistent kernel with Polars and Pandas support, which is nice for iterative analysis without re-authenticating. The caching layer makes the most sense if you're asking similar questions repeatedly or have a team that hits the same data from different angles. Works best when you actually feed it learnings rather than treating it as a one-shot query tool.