Hello all, i need some opinions regarding a project that im trying to evaluate should i build or not. I want to make a lightweight universal plug and play prompt optimizer that is only focused on search based prompt optimization. Akin to what dspy is doing with their teleprompters(Bootstrap, Bootstrapfewshot, MIPROv2, GEPA optimizers). why it would be a valuable endeavor(according to me):
- Single policy (only and only focused on search based prompt optimization)
Extremely extensible (implement your own algorithms easily)Plug and play, very simple for non developers and developers as well.will feature a cli based tool for on the go auto prompt optimization based on task and input data ( detect num of examples and autoselect the best prompting method unless specified)Very similar API to huggingface trainer for developers to quickly write up a prompt optimizer.compatible with hf datasets. Why i should not build this:
- I don’t know whether demand for such a thing exists or not and if people would just consider to use dspy.
DSPY already does what i tend to do but more things. this is what i am kind of thinking of: Chonkie is a library for chunking texts only, its very beginner friendly and plug and play for chunking texts, prior to that we always had recursive character text splitter, semantic splitter etc. from langchain and llamaindex with their own implementation. Yet chonkie is a well respected library. So i am really conflicted whether to build this or not. Please help. should i build this?