n8n-Based AI Knowledge System (Enhanced MVP)
├─ 1. Knowledge Ingestion Flow
│ ├─ Input → Clean → Structure
│ ├─ Embedding
│ └─ Store (knowledge_base)
├─ 1.5 Lightweight Distillation
│ ├─ Trigger (on insert OR batch)
│ │
│ ├─ AI Processing
│ │ ├─ summary (1~3 lines)
│ │ ├─ key_points (2~5 bullets)
│ │ └─ principles (optional)
│ │
│ ├─ Output
│ │ ├─ distilled_summary
│ │ ├─ distilled_points
│ │
│ └─ Store back into same KB record
├─ 2. AI Conversation Flow (RAG)
│ ├─ Query → Embedding
│ ├─ Vector Search (top_k=3~5)
│ ├─ Retrieval
│ │ ├─ content
│ │ ├─ distilled_summary
│ │ └─ key_points
│ │
│ ├─ Context Builder
│ ├─ AI Answer (with citation)
│ └─ Log
Summarize the following knowledge into:
1. A concise summary (max 2 sentences)
2. 3-5 key points
3. (optional) 1 general principle
Keep it factual and structured.
Return JSON:
{
"summary": "...",
"key_points": ["...", "..."],
"principle": "..."
}
.............................................
MVP Flow (Run-First Version)
├─ 1. Knowledge Ingestion
│ ├─ Input (Webhook / Manual)
│ ├─ OpenAI (clean text)
│ ├─ Generate embedding
│ └─ Save to DB (kb_id + content + vector)
├─ 2. AI Query Flow
│ ├─ User input
│ ├─ Generate query embedding
│ ├─ Vector search (top_k=3)
│ ├─ Build context
│ ├─ AI answer
│ └─ Return result
└─ DONE (no logging / no distillation / no scoring yet)