Hi Gustaf,
Looking forward to a joint effort. The C module is needed to produce the document embeddings. I'm on my way back from holidays. I'll post the details as soon as am back home. In short, I plan to use https://github.com/skeskinen/bert.cpp for the underlying library. This will allow me to load an existing huggingface model e.g. all-mpnet-base-v2. Storing and indexing with pgvector and solr should be simple. faiss might need its own TCL binding. We'll see.
I'll explain why this is a superior approach to plain search when I get home.
