Hacker News Reader

Just fyi, if you use a standard knn library like faiss and pretty much any embedding + language model raw from huggingface or an API, it will require ~15 lines of code to do what you describe. I’m not sure how much shorter langchain made your implementation but I can’t imagine it saving too much.

The point is in 99% of use cases no one will ever swap OpenAI or vector storage db for anything else. All Langchain does in these cases is introduces useless abstraction that takes longer to implement and makes things less transparent.

And when you need some customization instead of taking you 10 minutes it takes an hour to work around this abstraction.

Could you please share the code? I tried doing the same and, like the people in that thread, could not. TIA!