BioInsight: Multi-Agent Orchestration for Interactive Biomedical Knowledge Discovery

Biomedical researchers increasingly use language-model agents to interpret high-throughput protein association results, but existing systems often entangle retrieval, reasoning, writing, and visualization in opaque end-to-end pipelines. We present BioVoyager, a harness-level multi-agent orchestration framework for disease-centered biomedical evidence synthesis. BioVoyager introduces explicit artifact contracts that define the inputs, outputs, and validation criteria of each agentic stage, separating evidence retrieval, mechanistic reasoning, citation-grounded writing, and deterministic dashboard construction. Given a disease name, protein association table, and optional cohort metadata, BioVoyager coordinates pathway enrichment, disease-specific literature retrieval, evidence ranking, mechanistic synthesis, report generation, and dashboard assembly. By exposing intermediate artifacts that link protein signals to pathways, publications, mechanisms, and visual summaries, BioVoyager makes biomedical interpretation more controllable, auditable, and evaluable. Experiments and expert review show that artifact-contract-based orchestration improves citation grounding, evidence traceability, disease-specific synthesis, and interface reliability.