Cliniclaw:基于 AI 的 HIS 尝试,带有策略控制的临床智能体

1作者: chesterguan大约 2 小时前
功能:通过信任治理层(VERITAS)运行 AI 智能体(分诊、医嘱、实验室检查结果审核、药房、文档记录),用于临床诊疗。每个智能体的操作都受到策略控制(OPA Rego),以 FHIR R4 格式持久化存储,并进行加密审计。 关键设计决策: * 默认拒绝——智能体未经策略批准不得行动 * 人为干预的门槛由策略强制执行,而非用户界面约定 * FHIR R4 作为原生数据模型(无专有模式) * 加密审计链(SHA-256,仅追加) * 可插拔的 LLM 后端(Claude API / Ollama / mock) 技术栈:Rust, axum 0.7, tokio, regorus (Rego), sqlx, reqwest, Next.js 15 并非旨在取代 Epic——解决 Epic 在架构上无法用 AI 解决的问题。 演示:https://screen.studio/share/VjDVsjUG VERITAS github 仓库:https://github.com/Chesterguan/veritas
查看原文
What it does: Runs AI agents (triage, orders, lab review, pharmacy, documentation) through a trust governance layer (VERITAS) for clinical encounters. Every agent action is policy-gated (OPA Rego), FHIR R4-persisted, and cryptographically audited.<p><pre><code> Key design decisions: - Deny by default — agents cannot act without policy approval - Human-in-the-loop gates enforced by policy, not UI convention - FHIR R4 as the native data model (no proprietary schemas) - Cryptographic audit chain (SHA-256, append-only) - Pluggable LLM backend (Claude API &#x2F; Ollama &#x2F; mock) Stack: Rust, axum 0.7, tokio, regorus (Rego), sqlx, reqwest, Next.js 15 Not trying to replace Epic — solving problems Epic architecturally can&#x27;t solve with AI. </code></pre> Demo:https:&#x2F;&#x2F;screen.studio&#x2F;share&#x2F;VjDVsjUG<p>VERITAS github repo:https:&#x2F;&#x2F;github.com&#x2F;Chesterguan&#x2F;veritas