开发者在Hacker News上发布了ai·rete·RAG,这是一个将Rete规则引擎与RAG技术相结合的系统,用于实现可审计的AI决策1。该系统采用两阶段架构:纯Python编写的Rete引擎根据YAML规则生成决策,随后由RAG从策略文档中检索相关内容,并通过LLM生成决策的解释1。
系统支持嵌套规则(包括all/any/not逻辑)和前向链式推理,并能追踪完整决策过程1。审计模式下,系统会记录每条规则的评估情况,包括那些未被触发的规则1。规则可以引导检索范围,而检索到的文本则可转化为规则的输入事实1。
为降低使用门槛,平台提供了可视化编辑器供非技术人员操作,用户也可上传策略文档,由LLM自动生成规则草稿1。系统已推出免费层托管服务,包含MCP服务器集成功能1。开发者已在贷款、欺诈、临床、保险、法律、运营、电商和区块链等八个领域提供演示1。其中MCP客户端已开源(MIT许可),但引擎和平台暂未开源1。
A developer has unveiled ai·rete·RAG on Hacker News, a system that combines a Rete rule engine with Retrieval-Augmented Generation (RAG) technology to produce auditable artificial intelligence decisions 1. The platform employs a two-stage architecture in which a pure Python Rete engine generates decisions based on YAML-formatted rules, while RAG retrieves relevant content from policy documents and uses a large language model to generate explanations for those decisions 1.
The system supports nested rules using logical operators—all, any, and not—along with forward-chaining inference and decision tracking capabilities 1. An audit mode records the evaluation status of every rule, including rules that were not triggered, providing complete transparency into the decision-making process 1. Rules can guide the scope of document retrieval, and retrieved text can be converted into input facts for the rules 1. Non-technical users can leverage a visual editor to build rules, or they can upload policy documents from which the LLM generates rule drafts 1.
The platform offers a free tier of hosted service and includes integration with MCP servers 1. Eight demonstration domains are available, spanning lending, fraud detection, clinical applications, insurance, legal, operations, e-commerce, and blockchain 1. The MCP client has been released as open source under an MIT license, while the engine and platform remain proprietary 1. The developer has posed the question of whether others need assistance explaining automated decisions to regulators or auditors 1.
评论
还没有评论,欢迎留下第一条。