Cross-Domain Applications of LLM-Based Retrieval and Dialogue Systems: A Review of Current Practice
Ammoon Birzoim · 2025
Recent advances in large language models (LLMs) have enabled retrievalaugmented dialogue systems to support domain-specific information access with non-parametric grounding. This review surveys academic applications of LLM-based retrieval and generation systems across four domains: healthcare, education, finance, and law. We examine architectures combining dense retrieval and LLM decoding, assess their empirical performance in domainspecific tasks, and identify implementation variations such as vector stores, hierarchical retrieval, and knowledge graph integration. In healthcare, retrieval from clinical guidelines reduces hallucinations in diagnosis support; in education, systems leverage curriculum-aligned content for tutoring and fact explanation; in finance, document-grounded responses improve financial question answering and analytics accuracy; in law, structured retrieval ensures interpretability and traceability of case law outputs. We further analyze evaluation methodologies, domain-specific benchmarks, and alignment mechanisms. Our findings indicate a trend toward modular, citation-supporting architectures tailored to domain constraints, with domain-specific retrieval playing a central role in factual consistency.