hi all would appreciate some direction

MMitchMcD

hi all, would appreciate some direction on whether llama-index can help for my scenario:
say i have two manuals : how to fix a bike and how to fix a car. both are very long docts.
i ask a question about fixing a car.
i need an agent to figure out which manual to consult with -> consult with the appropriate manual -> give me the right answer.
can you pls recommend what i could employ for this scenario?
thx much!

12 comments

LLogan M

Yea, this is exactly what the router query engine does. We also have our own llama-index agents which should do a very similar thing (but with conversation history)

MMitchMcD

thank you Logan!

MMitchMcD

will go , learn about router query now

MMitchMcD

Logan, do I understand correctly the router engine redirects queries between list index and vector index, which are over the same data (docs in a paul graham folder)? My case is a bit different, I have two different files and will create two vector databases (since they are long).

MMitchMcD

do i create two vector indices , with different names, and just add in the "description" which index is useful for what?

LLogan M

The router engine can route between any number of indexes, and indexes of any type 🙂

So yea, you can create two vector indexes with different names and descriptions, and give those to the router query engine. Then, it should work 🙏

MMitchMcD

sorry , it's probably a very stupid question, but how is that different from langchain implementation https://python.langchain.com/docs/modules/chains/additional/multi_retrieval_qa_router ?

LLogan M

tbh it's probably pretty similar lol

Just that ours works with llama-index indexes (and also has the option to let the LLM select/query multiple indexes)

LLogan M

I think you probably found this notebook? But just making sure https://gpt-index.readthedocs.io/en/latest/examples/query_engine/RouterQueryEngine.html

MMitchMcD

yes, i did, thank you but got confused because it was creating an index over the folder and i needed two separate ones over each separate file

LLogan M

You can do seperate files with SimpleDirectoryReader, something like this, for group of file(s)

documents = SimpleDirectoryReader(input_files=["file.txt"]).load_data()

MMitchMcD

thx a lot!

Add a reply

Find answers from the community

hi all would appreciate some direction