AI Resonance · 阅读最新日报 · AI 入门推荐

2026-11-06 · America/Los_Angeles · 社区动态 · #9

I replaced vector-search RAG with Markdown + Knowledge Graphs using Claude CLI. The results surprised me.

I want to share something I've been experimenting with recently. Not claiming this is some revolutionary architecture or that I've solved RAG. I'm just curious if anyone else has tried something similar. I was building an AI system for a fairly complex ERP application. Initially, I used the usual RAG approach: embeddings, vector search, retrieving relevant chunks, and passing them to the LLM. But honestly, I wasn't happy with the results. The problem wasn't always the LLM. Sometimes the retrieved information itself was incomplete. The model could understand individual database tables, but…

I replaced vector-search RAG with Markdown + Knowledge Graphs using Claude CLI. The results surprised me.

热度 46.0 / 100;排名与评分保留该期记录。