Personal Knowledge Base (RAG)：把链接和帖子沉淀成可检索知识库

• A folder-based corpus plan (notes/, links/, summaries/) is prepared.
• Choose an embedding/retrieval backend and decide refresh cadence.

1. Ingest URLs and social posts with metadata: source, author, date, topic.
2. Chunk documents by semantic section, not fixed length.
3. Create a retrieval prompt template that asks for citations in every answer.
4. Run weekly dedup + quality pass to remove stale/low-signal chunks.

• Without metadata, retrieval quality drops sharply when corpus grows.
• PII handling policy should be documented before importing private notes (needs verification).

This tip is aggregated from community/public sources and preserved with attribution.