跳转到内容

Gold Standard — 18 Lessons Learned(黄金标准:经验教训)

知识库位置:FDE Knowledge Base/06 Case Studies/Reference-Projects/01-Enterprise-Knowledge-Assistant/18-Lessons-Learned.md

Enterprise RAG is partly a knowledge governance problem. (企业 RAG 部分上是知识治理问题。)

Promote to(晋升至):

Knowledge Assistant Pattern(知识助手模式)


Permission filtering must happen before retrieval context reaches the model. (权限过滤必须发生在检索上下文到达模型之前。)

Promote to:

ARCH-001 Enterprise RAG Architecture(企业级 RAG 架构)


Transactional state should remain in APIs / databases. (事务状态应留在接口/数据库中。)

Promote to:

Decision Guide:RAG vs Direct Database Query(RAG vs 直接数据库查询)


Exact identifiers require lexical search signals. (精确标识符需要词法检索信号。)

Promote to:

RAG Retrieves Wrong Documents(RAG 检索错文档)


Evaluation must include(评估必须包含):

No-answer(无答案)
Restricted(受限)
Stale(过期)
Exact-code(精确编号)

not just happy-path semantic questions.(而不只是顺利路径的语义问题。)

Promote to:

RAG Evaluation Toolkit(RAG 评估工具)