Rubric-Based Evaluations to Improve LLM Retrieval and Search Functionality