Skip to main content

Architecture Overview

Clinical Corvus is a full-stack clinical decision-support platform designed for workflow utility, privacy, and auditability.

Top-Level Components​

ComponentResponsibility
Clinical Workspace (UI)The clinician-facing cockpit for rounds, handoff, and structured drafts.
Orchestration APIReceives requests, builds context, routes work across agent roles, and returns reviewable outputs.
Retrieval & EvidenceHybrid retrieval over curated content, producing traceable context packs and citations.
Clinical StateA structured, auditable representation of the active clinical session (episode-level).
Tool Gateway (Optional)Controlled access to external sources for evidence retrieval under privacy rules.

What Makes This Architecture "Clinical-Grade"​

  • Workflow-first outputs: plans and handoffs are returned as drafts that teams can edit.
  • Bounded autonomy: the system is designed to avoid silent execution; review is explicit.
  • Local-first privacy posture: patient-identifying context is treated as sensitive by default.
  • Audit-friendly evidence: when evidence is requested, results aim to include citations that can be checked.

Self-Reliant Infrastructure​

Clinical Corvus implements a cost-effective, self-reliant infrastructure for intelligence gathering that minimizes dependency on paid APIs while maintaining high-quality evidence retrieval.

Strategic Goals​

  1. ✅ Eliminate recurring API costs for routine searches
  2. ✅ Establish local medical corpus for high-yield clinical content
  3. ✅ Implement tiered routing with strict budget controls
  4. ✅ Ensure operational resilience through self-hosted components

Tiered Search System​

The infrastructure uses a three-tier search system with automatic escalation:

Tier 1 (Green) - Always Free​

  • Sources: Local medical corpus, PubMed, Europe PMC, OpenAlex, Crossref
  • Cost: $0
  • Latency: 1-3 seconds
  • Use Case: All routine clinical queries

Tier 2 (Yellow) - Self-Hosted​

  • Source: Self-hosted metasearch aggregator
  • Cost: $0 (self-hosted)
  • Latency: 2-5 seconds
  • Trigger: Tier 1 returns insufficient results

Tier 3 (Red) - Paid APIs​

  • Sources: Premium search APIs
  • Cost: ~$0.005-0.01 per query
  • Latency: 1-2 seconds
  • Trigger:
    • Explicit user opt-in, OR
    • High-recency query detected (keywords: "latest", "newest", "recent")
    • Monthly budget: $0 by default (must be explicitly enabled)

Medical Formulas Available​

The system includes 55+ offline medical formulas across clinical domains:

Sepsis & Critical Care​

  • qSOFA Score, SOFA2 Score, MEWS, NEWS2, LODS

Cardiovascular​

  • CHA2DS2-VASc, HAS-BLED, GRACE Score, TIMI Risk, HEARTS Score

Pulmonary​

  • Wells Score (PE/DVT), Geneva Score, BAP-65

Renal & Metabolic​

  • CKD-EPI, MDRD, Cockcroft-Gault, MELD, MELD-Na

Hematology & Oncology​

  • CLARIFY Risk, Gleason Score, IPSS

Gastrointestinal​

  • Child-Pugh, FIB-4

Pediatric​

  • Pediatric Glasgow Coma Scale, Pediatric Appendicitis Score

Success Metrics​

MetricTargetNotes
Monthly API costs$0-5/monthReduced from ~$50/month baseline
Local corpus coverageTarget: 80%Of routine clinical queries
Search latencyTarget: 1-2sImproved from 2-3s baseline
Result relevanceTarget: 85%Improved from ~70% baseline
note

Metrics are internal targets based on development testing. Actual performance may vary based on query types and corpus content. Independent validation is planned.