Find answers from the community

Updated 2 months ago

Handling and queries in RAG pipeline

Hey guys, any idea on how exactly you respond on queries in RAG pipeline that use the word and ? example: Find me presentations related to LLMs and AWS Should I have a query rewritier specifically to split queries these type of queries and then only consider common contexts retrieved by the split queries
W
t
2 comments
I think this query should show similar content to LLM and AWS regardless of any change if they exist in your docs.


If it is not coming and you wanna segrate queries like check if there are two queries and wanna ask them separately then you can simply keep one layer of llm call to identify this case and then ask it from there onwards based on the number of queries.
I have the idea of using spacy for entity recognition, classification and their relationships

We extract 2 things from the user query
  1. Entitys - AWS, LLM
  2. Operators - and, or, greater less, less than etc...
Once these are classified then we can put all the different use cases logic and perform vector search.

I am not sure if this is even feasible in RAG pipeline.
Add a reply
Sign up and join the conversation on Discord