Neat examples, and I agree that extending SQL like this has real potential. Another project along very similar lines is https://github.com/ryrobes/larsql
I work on https://github.com/xtdb/xtdb which is broadly Postgres-compatible with a few key SQL extensions (SQL:2011 bitemporal tables + immutability, first-class nested data, pipeline syntax, etc). Built on Arrow and the JVM but is otherwise mostly from scratch.
XTDB is perhaps not directly relevant to the topic at hand, but I am a firm believer that ML workflows can benefit from robust temporal modelling.
Someone shared with me these the other day and we're inspired to add more remote LLM calls directly into ShapedQL now: https://github.com/asg017/sqlite-rembed https://github.com/asg017/sqlite-lembed