Multi-agent systems work best when each agent has a narrow, verifiable responsibility. A central runtime coordinates literature retrieval, reasoning validation, and citation integrity as separate concerns.
Literature retrieval agent: ArXiv, Semantic Scholar, PubMed integration.
Reasoning validation agent: rule-based checks on evidence-backed outputs.
Citation integrity agent: cross-reference validation before report generation.