How to Analyze Tables In Large Financial Reports Using GPT-4 (w/Jerry Liu, LlamaIndex)

In this video, you'll learn how to ask complex questions on tables embedded within large text using GPT. In this presentation we use Openai's latest `gpt-4-1106-preview` model.

The challenge for performing simple RAG (retrieval augmented generation) is that text splitting may break up tables and similarity search is likely to fail. To overcome this problem, we store raw tables and text along with table summaries better suited for retrieval. The tool used to orchestrate and enable this solution is LlamaIndex: a data framework for LLM applications to ingest, structure, and access private or domain-specific data.

During this presentation, Jerry Liu, Co-Founder of LlamaIndex, covers the use case of two "semi structured" data sources: 1) A large wikipedia page containing tables of billionaires 2) 200+ pages of Tesla's 10-k financial report and discusses the strategy with Mayo Oshin.

-----------------------

LlamaIndex links:-

Chat with data links:-

Timestamps:

01:30 Wikipedia article demo and architecture
05:19 Codebase walkthrough
19:45 Analysing Tesla 10-k filing
30:15 Question and answer session

#openai #gpt #finance #investing

Hey, thanks for this video! Amazing content!
One question: what if the tables go over other pages, but the header only appears on the first page?

[lavamonkeymc]

How does the complex document hierarchical node-based index work with vector DBs? I’ve found that Azure Cognitive Search ( very popular in industry where data is private ) does not work with these node models. Are there specific vector dbs that work with this?

[sanchitshaleen]

Would the above example notebook work using open-source LLMs like the Llama's as well?

[zohaib]

at 11:51 he said he will describe a way to generate table summaries automatically later in the video. But then it was never shown, or did I miss something here ?

[chatwithdata]

Timestamps:

01:30 Wikipedia article demo and architecture
05:19 Codebase walkthrough
19:45 Analysing Tesla 10-k filing
30:15 Question and answer session

[heatvisuals]

Can it extract the data from my wells fargo savings statement report? I have tried everything and the wells fargo statement report is too dynamic lmao.