Using the RAG feature of CALM

In the paper for CALM, I notice that the bot can trigger a command KnowledgeAnswerCommand and query a database for answers like in RAG.

How do I input my own data into this database and how do I customize how the database operates (customize chunk sizes, embedding model, minimum relevancy score, etc.). Afterwards, how do I instruct the bot to query a specific database at a specific user command

@BaoTranHuyDuc Please read the docs for Enterprise Search:

OOTB - rasa uses FAISS as the vector DB. If you use the OOTB functionality, it is important to know:

  • it only accepts text files
  • customisation is limited. For example, you currently cannot customize chunk size, relevancy score. You can choose the embedding model by setting it in the config.yml (see here for more details).

If you want to customise the ETL of the source data more, you will have to build your own ETL pipeline and use an external vector DB.