Find answers from the community

Updated 2 months ago

Hello everyone,

Hello everyone,

I have a question that I hope some of you might be able to help with. We're currently building an intelligent chatbot for a client (a governmental entity) aimed at assisting entrepreneurs with setting up their businesses, financing, etc. So far, everything has been going great, and our Retrieval Augmented Generation (RAG) system, built with LlamaIndex, allows our LLM to provide extremely relevant solutions. However, we've encountered an issue regarding the display and retrieval of LINKS.

The bot consistently proposes the right solution, but often the associated link is a hallucinated URL. When we use only about 50/60 links, the results are perfect. But as soon as we expand to using 500 links, we keep running into this issue.

Does anyone have any ideas on how to optimize URL retrieval to exclusively use the URLs provided in our database?

Thank you in advance for any insights or suggestions you might have! Have a good day πŸ™‚
W
T
L
4 comments
I think providing some contextalong with the URL would help if it is not being done already

URL: abc.gov.in
Context: This url helps to perform this operation and allows these privilages.... etc


That way correct urls will be retrived IMO
thank you a lot for your reactivity! I'm going to try it and come back to keep you informed.
In our case we added url as document metadata, and using query pipelines you can parse that metadata
oh, interesting! i'll try it later. Thanks!
Add a reply
Sign up and join the conversation on Discord