Use cases for data pipelines with LLMs and Structured data sources

[anupsurendran]

This was initially posted in a reddit discussion below:https://www.reddit.com/r/dataengineering/comments/156ltst/use_cases_for_llms_on_structured_data_sourced/

Does anybody have use cases around data pipelines for LLMs and structured data sources? If so how do you deal with the following :

Context mismatch - since enterprise databases have very specific schemas and terminology that the LLM is unaware of, how do you prevent misunderstanding?

Entity ignorance - entities are a relational db terminology but how do you get that into the LLM world?

Compositionality - LLMs lean towards a holistic view, and our database queries require hierarchical, compositional views. How do you deal with this?