Still iterating on Trilogy, a SQL variant with an embedded semantic layer that removes the need for joins and better typing/functions.

Spent some time last month improving array handling + error messages and UX and adding an MCP server option; Claude does pretty well already but there's some syntax/error tweaks to make it simpler for it and humans.

Then pivoting back into scheduling + materialization optimizations (identify common aggregates across several scripts and automatically build the common datasets for reuse).

Link: https://github.com/trilogy-data/pytrilogy