Idea

This is a continuation of work done by Beekeeping advice with LLM.

Next step of this feature is Audio-interface (to talk with bees)

We want here to not only give advice based on hive state, but also allow user to do question-and-answers. For such querying we need to make LLM smarter by also building own knowledge base.

Knowledge base initially could be just a

• dataset of beekeeping books [need to collect available books though]

Over time we could build knowledge base from some internal, discovered data or patterns, like bee movements, sounds etc.

Which means we need to

• take knowledge base
• encode it sentense by sentense and pages as embeddings
• store embeddings in vector DB
• on query, do an embeddings similarity search, retrieve closest topics
• enrich context, pass to LLM, generate output

In terms of UX, we can make LLM act such that it would impersonate a bee colony