This approach has really helped me out in my work. I do something very similar using DuckDB to slurp output files anytime I write a custom hierarchical model. The single sql queryable file simplified my storage and analytics pipeline. I imagine SQLite would be especially ideal where long term data preservation is critical.