Looking at building a chatbox implementation for our forums so that common questions that repeat can be done via a simple query interface. If I can find and implement a test case before the server move, it may dictate the resources I allocate when that move is complete (GPU's, ram etc).
Right now I'm looking at LlamaIndex Query Pipelines with Text-to-SQL as my first attempt, and will expand to other solutions after that, but I want to remain open source and self hosted rather than reliant on ChatGPT or other things that will have to be maintained as they update their services. This shouldn't be a high overhead project, but just an intro to building custom RAG/DAG datasets for existing chat models.
Open to input from anyone who is familiar with this subject.