TRACE converts user queries into structural and semantic constraints via a dual data representation schema to improve recommendation accuracy in noisy service directories. The framework mitigates hallucination in public service chatbots by parsing constraints for retrieval rather than relying on general-purpose LLM web knowledge.