华盛顿大学医学院发布 WUEM Sim Intel v3.9.0，用 AI 管理临床模拟中的潜在安全威胁

WUEM Sim Intel

WUEM Sim Intel is a simulation improvement and Latent Safety Threat (LST) management system developed for the Washington University School of Medicine, Department of Emergency Medicine.

The platform helps simulation teams turn high-fidelity clinical scenarios into usable safety work: structured post-session reports, reliable LST identification, follow-up tracking, searchable institutional learning, and documentation that can be shared with educators, safety leaders, and operational stakeholders.

Current release: v3.9.0 — provider-flexible AI report drafting, case-summary context, a redesigned clinical workspace, persistent dark mode, and a hardened Cloudflare deployment path.

🏛️ Project Overview

WUEM Sim Intel is specialized for clinical simulation programs that need more than a written debrief summary. It supports a Just Culture approach by helping teams document what happened, identify system vulnerabilities, assign follow-up, and preserve lessons learned across repeated simulations.

The system is designed around the practical workflow of a simulation program:

Capture what occurred during a session.
Generate a polished post-session report.
Identify and classify LSTs.
Track mitigation status and institutional follow-up.
Search across prior sessions, scenarios, and safety findings.
Export clean documents with supporting photos and media.

Key Capabilities

🏥 LST Identification & Management: Centralized tracking for Latent Safety Threats with categories, severity, status, mitigation notes, and institutional audit history.
📋 Simulation Report Generation: Converts session notes into professional reports grounded in simulation education, safety improvement, and Just Culture principles.
✅ Follow-Up Workflow Support: Keeps safety findings visible after the session so unresolved threats can be reviewed, escalated, and closed.
🔍 Institutional Learning Search: Search across prior scenarios, reports, and safety findings to recognize repeated patterns and recurring operational risks.
📸 Session Media Documentation: Attach simulation photos and high-resolution media, then include them in exported reports as polished photo collages.
🤖 Provider-Flexible Clinical Assistant: Uses OpenAI or Gemini for report drafting and LST extraction, with retrieval over prior simulation evidence to support faster synthesis.
🧾 Case-Aware Drafting: Includes concise case-file summaries in the report prompt so generated drafts reflect the selected scenarios without overwhelming the model context.
🌗 Accessible Clinical Workspace: A responsive clinical-editorial interface with shared, persistent light and dark themes across the simulation suite.

🏗️ System Architecture

Built on a Cloudflare-native stack for reliability, fast access, and secure clinical education workflows:

Frontend: React (Vite + TypeScript + Tailwind) deployed to Cloudflare Pages at https://intel.wuemsim.org.
Backend API: Cloudflare Workers (Hono) running at the edge. The public workers.dev route is disabled, and the Pages Function proxy calls the Worker through the WASHU_SIM_INTEL_API service binding.
Decision Support:
- Library Q&A: Retrieval-augmented answers over the local simulation knowledge base.
- Report Drafting: Provider-agnostic asynchronous streaming via the OpenAI Responses API or Google Gemini.
- Semantic Indexing: Cloudflare Vectorize with Workers AI for retrieval across scenarios, reports, and LST records.
Data Primitives:
- Relational SQL: Cloudflare D1 (washu_sim_db)
- Object Storage: Cloudflare R2 (washu-sim-intel-storage-2026)
- Metadata Cache: Cloudflare KV (RATELIMIT_V2)
- Vector Index: Cloudflare Vectorize (sim_search)

🛠️ Getting Started

Prerequisites

Node.js: v20 or higher.
Cloudflare Account: With access to D1, R2, Vectorize, Workers AI, KV, Pages, and Workers.
Secrets:
- OPENAI_API_KEY: Preferred provider for report generation, library Q&A, and LST extraction.
- GEMINI_API_KEY: Optional fallback provider.
- TURNSTILE_SECRET_KEY: For spam protection on generation endpoints.
- ADMIN_TOKEN: For protected clinical data and administrative API access.

Local Development

Clone the repository:

git clone https://github.com/salthepal/WashUSimIntelligence.git
cd WashUSimIntelligence

Frontend Setup:
```
npm install
npm run dev
```
Backend Setup (Wrangler):
```
cd worker
npm install
npx wrangler dev
```

🚀 Deployment

The system uses GitHub Actions for continuous delivery:

Frontend: Automatically deployed via Cloudflare Pages and served at https://intel.wuemsim.org.
Backend: Update resource IDs in worker/wrangler.toml and run npm run deploy in the worker/ directory.
Proxying: Production /api/* traffic flows through the Pages Function service binding WASHU_SIM_INTEL_API. Use an explicit BACKEND_URL only for local or break-glass testing.
Access: Cloudflare Access protects intel.wuemsim.org, washusimintelligence.pages.dev, and Pages preview hostnames for wustl.edu users plus the configured admin email.

See DEPLOYMENT.md for resource bindings, secrets, migrations, and deployment checks.

📦 Releases

Release notes and version history are available in GitHub Releases and CHANGELOG.md.

🔒 Security & Governance

Just Culture: Reports prioritize growth and systemic fixes over individual blame.
Data Residency: Current production placement is Cloudflare D1 in ENAM and R2 in WNAM. Treat that as approved only for non-PHI simulation safety data; migrate to newly created jurisdiction-pinned resources before accepting stricter residency requirements.
Administrative Access: Clinical data reads and writes require X-Admin-Token.
Spam Protection: Upload and generation endpoints are protected by Cloudflare Turnstile.
Security Headers: Browser protections are served from public/_headers.
Clinical Data Handling: Avoid entering patient identifiers or protected health information unless your Cloudflare deployment has been approved for that use.

📜 License

This project is licensed under the GNU General Public License v3.0 (GPLv3). See LICENSE.md for full details.