Sounds very similar to a databricks style datalake, except with a custom file format instead of parquet.