LlamaIndex Sessions: Practical challenges of building a Legal Chatbot over your PDFs

In this video, we chat with Sam Yu on practical challenges of 1) parsing supreme court decisions, and 2) building an LLM-powered chatbot over it.
A lot of challenges in building a retrieval augmented system boil down to challenges in parsing the data. We talk about different strategies for parsing, the pros/cons of different PDF parsing/OCR packages, and also different retrieval strategies.
Background:
Sam is an AI product engineer currently developing an application with AI capabilities. His goal is to utilize a LLM in order to democratize specialized domain knowledge, making it accessible to everyone.

[@kevon217]

That pdf discussion was super helpful, thanks!

[@vigneshpadmanabhan]

Amazing session. Much appreciated if you could also include his profile links.

[@ThumanShield]

Are there any links available for Sam’s work?

[@walkaman4396]

What's the link too? 😊

[@joser100]

Any thoughts on the area of extracting relevant data from diagrams in PDFs? maybe not very relevant to this field, but maybe there are also some important diagrams that may appear in those legal docs as well...

[@jewpaltz]

This was very interesting.
Where can we find more information on the different products that you tied and that you mentioned in this video?
I am particularly interested in extracting tabular data from tables in PDFs and then feeding them to LLMs to answer questions about the data.
Where did you see good comparisons between methods of parsing PDFs?

[@juliakoncha5988]

I once tried Unstructured Library, which is tesseract-based. It did a good job, but my table had a pretty easy structure:)

[@kevon217]

Hybrid search is very effective I’ve found and adds a nice layer of customization