Find answers from the community

Updated 7 months ago

Hi there, I used `llama-parse` and

Hi there, I used llama-parse and implemented the RAG on a set of financial documents. Similar to the example in this notebook[1], we build a raw index and recursive index . To my surprise, the results from raw_index.as_query_engine are more accurate than the recursive one. I am try to get an intution for why this might be? For context, we have tables with financial data and a sample query might look like - what was the total rent for Property A in 2023? What is the key difference between the two indices? and how do they work under-the-hood?

https://github.com/run-llama/llama_parse/blob/main/examples/demo_advanced.ipynb
k
L
2 comments
More on this, here are the responses i get:

**New LlamaParse+ Basic Query Engine**
The Total Payroll and Related Expenses at xyz for the period Oct 2022-Sep 2023 were $1,200.
Retrieval entering a00568c4-6f41-4410-a080-a31b16a61d3e: TextNode
Retrieving from object TextNode with query What were the Total Payroll and Related Expenses at xyz ?

**New LlamaParse+ Recursive Retriever Query Engine**
The total payroll and related expenses for the year were $500. The exact location xyz is not mentioned in the provided context.

The Basic Query Engine seems to retrieve the relevant context but the recursive engine doesn't. I wonder why this might be πŸ€”
Hmm, not sure tbh. I know we GREATLY improved the text mode on llamaparse recently
Add a reply
Sign up and join the conversation on Discord