Unsatisfactory results with a resume PDF

[sameer.s.mahajan]

First of all thanks for this wonderful course. As a custom use case I was trying it with a simple resume. I used Azure Open AI, text-embedding-ada-002 for embeddings, TokenTextSplitter, chroma vector DB. For chat model I tried GPT 3.5 turbo as well as GPT4. However it could only answer

“What is this person’s name?” satisfactorily.

It could not answer any of:

“which cities?”
“what are universities in this document?”
“what is profile summary?”
“what is this person’s education?”
“what are educational details?”

Any ideas?

[sameer.s.mahajan]

Interestingly when I follow the approach suggested in the other course of ‘LangChain for LLM Application Development’ leveraging Q&A over documents, it works great!