Hi! Long time lurker, first time poster. Apologies, for reasking this. I work at a bank as a C# dev. We are currently transitioning from the MT Message Format to the new MX Format. This is for SWIFT messages. I wanted to train an LLM to help us field questions on the new format, as well as aid in coding. It would be like a consultant. Would I finetune a model or use embeddings?
You could use both, but I think embeddings would be a requirement in either case, and certainly the way to start out.
Get your documentation embedded and then see how the retrievals work for your use case, build up your familiarity with the performance and see if what you have works well enough, fine tuning is usually a lot of additional work.