Perfect.. I've seen so many demos showing people feeding data to LLMs so they can "ask questions about their data" but still not seen any real business use cases. Is anyone using these tools in production for a real problem?
The enterprise search domain has evolved quite a bit, utilizing & integrating retrieval-rerank architectures for a few years now. That includes embedding/vector storage, as well as LLM usage (eg using Google T5 for query understanding and model rerank)… there’s some generative modeling but not really like GPT… more like structuring an abstract tree from a complex Boolean query, and sending that tree into an custom retrieval approach over ANN & rerank ensemble.
If you’re doing that type of work in a serious manner you don’t use libraries like langchain or lamaindex. They are a bit late/irrelevant to the engineering that exists in that type of environment.
