Is anyone working on a personal assistant using RAG and LLMs?

joelthelion@lemmy.world to Programming@programming.dev – 32 points –

A personal chatbot that can search through your notes and saved documents sounds like an obvious application of LLMs, but for some reason I can't find such a project. Are there any in the works?

5

You can easily build one yourself. Check llamaindex and langchain for prebacked solutions. Otherwise the math is pretty trivial if you are using normalized embeddings. You can quickly do one with numpy.

My suggestion anyway is to start with langchain, full of tutorials executable as jupyter notebook

I did this with my wiki in order to read each article and come up with a category and keyword cloud. I first started by attempting to use many different tools to build my own LLM and then realized that I was reinventing the wheel and shifted to using OpenAI’s API. I was able to prompt it to read each article as input and output what I needed in a machine readable format. It worked like a charm.

If you're in the Google ecosystem, Bard recently gained the (toggleable) ability to access you email/notes (can't remember if it can access documents)