[LocalLLaMA] Create a library to reduce overhead in Langchain and Langgraph

I am wondering what does the community think about langchain and langgraph. Currently the organisation that I work for uses Langgraph and langchain in production applications for chatbots. The problems that I see, is langchain has more contrbutions and unneccesary codes, libraries coming in. Example: we use it only as inference but, pandas is also installed which is completely not necessary for my use case, pdf splitter is also not necessary for me. It has 3 to 4 ways of creating react agents or tool calling agents. This results in larger Docker image. We have invested in a different monitoring system and only use langgraph for building the graph and running it in a streaming scenario. I was wondering, if I can create a library with only the stuff that I use from langgraph and langchain, I will be better off without extra overhead. Even though we build multiagent workflows, I dont think langgraph will truly be useful in that case, given that it comes with Pre built prompts for the create\_react\_agent etc. Please let me know your views on the same.