Domain-Specific Query Understanding for Automotive Applications: A Modular and Scalable Approach

Collection of recent arXiv papers (Apr 21 2026) covering advances in LLM-driven and retrieval-augmented systems and datasets: a modular two-step query-understanding system for automotive applications; RAG-DIVE, a dynamic multi-turn evaluation framework for RAG; FlexStructRAG for multi-granular, structure-aware relational retrieval; MARA for multimodal adaptive RAG in document QA; Paper2Data and the UrbanDataMiner portal extracting global urban datasets from literature; a diagnostic study of LLM-based rerankers in cold-start recommender systems (coverage/exposure failures and mitigations); SciF