I wrote a tutorial for invoking Mistral 7B model directly with SQL using Python UDFs in Exasol and Ollama. This demonstrates a fully self-hosted AI pipeline where data never leaves your infrastructure—no API fees, no vendor lock-in. Takes ~15 minutes to set up with Docker.