Agents

Clinical Corvus uses specialized agent roles to produce outputs that are clinically usable, reviewable, and safe-by-design. This page describes the concept—not the proprietary implementation.

How A Request Becomes An Answer

1. Intake: the request is cleaned and scoped (what the user is trying to do).
2. Context snapshot: relevant patient/workspace context is assembled into a compact, consistent view.
3. Routing: the system decides whether the request needs:
   • a workflow draft (rounds/handoff/plan),
   • evidence retrieval (guidelines/literature),
   • clarification (missing details).
4. Reasoning: a clinical synthesis and draft plan are produced in a clinician-friendly format.
5. Evidence (optional): when requested, evidence is retrieved and summarized with citations.
6. Verification: outputs are checked for missing support, contradictions, and unclear risk framing.
7. Response shaping: results are formatted as drafts the clinician can accept/edit/discard.

Core Roles

• Clinical reasoning role: synthesizes the case and drafts plans aligned with acute-care workflows.
• Research role: retrieves and summarizes external evidence when asked, returning auditable citations.
• Verification role: acts as a safety gate before anything is shown to the clinician.

Conceptual Diagram

Guarantees

• Clinician-in-control: the system proposes; clinicians accept/edit/discard (check-to-accept).
• Evidence on demand: when evidence is requested, answers aim to include citations for audit.
• Workflow-first formatting: outputs are shaped for rounds, handoff, and action lists—not generic chat.