Skip to content

WISE_AI_CDSS_Architecture.md

Latest commit

 

History

History
192 lines (140 loc) · 4.81 KB

WISE_AI_CDSS_Architecture.md

File metadata and controls

192 lines (140 loc) · 4.81 KB

WISE AI — Agentic Clinical Decision & Continued‑Care System (CDSS)

Purpose: This document is the single source of truth for
architecture, scope boundaries, and demo behavior for the TSAI Capstone.

0. One‑Line Vision

WISE AI is an agentic Clinical Decision & Continued‑Care System that augments patients and doctors by synthesizing patient context, research signals, and platform intelligence — delivering explainable recommendations, clinical assistance, and continuous‑care support.

1. System Positioning

WISE AI operates as a standalone AI system that overlays existing EHR workflows (WISE Doctor) without tightly coupling to them.

• No EHR core refactor required
• Consent‑driven, user‑initiated invocation
• API‑level or browser‑level integration only
• Production‑safe by design

2. Core Components

2.1 CDSS Web Application (Primary UI)

A standalone web application that:

  • Accepts curated patient/doctor context
  • Synthesizes signals using agentic reasoning
  • Displays outputs with explanations & confidence
  • Showcases stretch actions (future‑flagged)

Outputs:

  • Patient recommendations (diet, diagnostics, escalation)
  • Doctor assistance (clinical notes, diagnosis hints, safety flags)
  • Confidence scores & feedback loop insights

2.2 Browser Plugin (Data & Research Adapter)

A user‑triggered extension that:

  • Extracts consented on‑screen data from EHR views
  • Parses PDFs, vitals tables, structured forms
  • Performs live research (WWW + curated sources)
  • Writes to the shared knowledge bank

No background scraping
No silent monitoring
Explicit user action only

2.3 Shared Knowledge Bank

Logical layers:

  1. Public research (WWW, guidelines)
  2. WISE platform intelligence (anonymized)
  3. Doctor workspace (optional contribution)
  4. Patient‑specific context (highest priority)

Used by:

  • CDSS reasoning agents
  • Research agent (plugin)
  • Confidence feedback loop

3. Agentic Architecture (Conceptual)

  • Orchestrator Agent
  • Context Synthesis Agent
  • Research Agent
  • Clinical Reasoning Agent
  • Safety & Explainability Agent
  • Confidence Scoring Agent
  • Action Simulator (future‑flagged)
  • All agents operate under a central Orchestrator with deterministic execution for demo

Agents collaborate but do not act autonomously in MVP.

4. Invocation Model

  1. User clicks “WISE AI” button in EHR
  2. Plugin captures consented context
  3. Data written to Knowledge Bank
  4. CDSS Web App opens in new tab
  5. Agentic reasoning executes
  6. Outputs rendered with sources & confidence

5. MVP Guardrails (Hard‑Locked)

  • No background processing
  • No automatic EHR writes
  • No autonomous actions
  • No medical claims
  • Human‑in‑the‑loop always

6. Stretch Goals (Demo‑Only, Simulated)

  • Send clinical note back to EHR (preview)
  • Trigger follow‑up reminders (preview)
  • Suggest lab orders (preview)
  • Compliance tracking dashboard (preview)

Clearly labeled: “Future Capability”

7. Why This Architecture Works for Capstone

  • Demonstrates real agentic design
  • Meets TSAI explainability expectations
  • Shows production realism without risk
  • Scales cleanly post‑capstone

8. Architecture Diagram (Rendered)

WISE AI CDSS Architecture

9. Architecture Diagram (Mermaid – Source)

graph TD

  %% Actors
  Patient["Patient User"]
  Doctor["Doctor User"]

  %% EHR
  EHR["EHR UI<br/>(WISE Doctor)"]

  %% WISE AI Components
  Plugin["WISE AI Plugin<br/>(Consent-based Capture & Research)"]
  CDSS["WISE AI CDSS Web App<br/>(Agentic Reasoning UI)"]
  KB["Shared Knowledge Bank"]

  %% Agents
  SymptomAgent["Symptom Agent"]
  CBCAgent["CBC / Lab Agent"]
  TrendAgent["Trend / History Agent"]
  ResearchAgent["Research Agent"]
  ActionAgent["Action / Recommendation Agent"]

  %% User Interaction
  Patient -->|Uses EHR| EHR
  Doctor -->|Uses EHR| EHR

  %% Invocation
  EHR -->|Click WISE AI| Plugin

  %% Data Capture
  Plugin -->|Consent-based Extraction| KB
  Plugin -->|Triggers Analysis| CDSS

  %% Agentic Reasoning
  CDSS --> SymptomAgent
  CDSS --> CBCAgent
  CDSS --> TrendAgent
  CDSS --> ResearchAgent

  SymptomAgent --> ActionAgent
  CBCAgent --> ActionAgent
  TrendAgent --> ActionAgent
  ResearchAgent --> ActionAgent

  %% Outputs
  ActionAgent -->|Recommendations and Guidance| CDSS

  %% Feedback Loop
  CDSS -->|Confidence Signals| CDSS
  CDSS -->|Optional Feedback| KB
Loading

10. Status

This document represents the frozen architecture baseline for:

  • MVP feature freeze
  • Capstone demo
  • Phases 0–2 execution

Any changes require explicit review.