A Generative AI Engineer is building a chain to take user input, format it to match a legacy system, pass it to an LLM, and then post-process the result before returning it to the application front end. The engineer wants to package this logic into a single deployable model. Which approach best supports this requirement?