Author narratives on RAG, MCP, knowledge bases, and traceable retrieval.https://langhuan.dev/enhttps://langhuan.dev/en/blog/chinese-fts-zero-recall/https://langhuan.dev/en/blog/chinese-fts-zero-recall/An everyday question like "埃及有哪些民族？" returns literally nothing from SQLite FTS5 Chinese full-text search — and hybrid retrieval quietly degrades to vector-only, with no errors, no alerts, and normal-looking metrics. The mechanism, the query-side fix, and how to check whether your FTS channel is already spinning in neutral.Tue, 25 Aug 2026 00:00:00 GMThttps://langhuan.dev/en/blog/retrieval-eval-benchmark/https://langhuan.dev/en/blog/retrieval-eval-benchmark/The metric that matters most for a knowledge base product is retrieval quality, and all we had were functional tests. How Langhuan built a retrieval eval with 200 human-annotated real queries, a dual-track channel matrix, and bit-identical reproducibility — plus the production bugs it caught on day one and the hybrid-search hypothesis it finally confirmed with data.Tue, 25 Aug 2026 00:00:00 GMThttps://langhuan.dev/en/blog/local-rag-knowledge-base/https://langhuan.dev/en/blog/local-rag-knowledge-base/A local RAG knowledge base does not need a full database and queue stack on day one. Start with real documents, Chinese-aware retrieval, and evidence; move to production components when the workload earns them.Tue, 18 Aug 2026 00:00:00 GMThttps://langhuan.dev/en/blog/mcp-knowledge-base-for-agents/https://langhuan.dev/en/blog/mcp-knowledge-base-for-agents/The second agent changes the question. Instead of uploading the same documents again, decide who owns versioning, retrieval, access, and source evidence. That is where an MCP knowledge base belongs.Tue, 18 Aug 2026 00:00:00 GMThttps://langhuan.dev/en/blog/project-knowledge-context-limits/https://langhuan.dev/en/blog/project-knowledge-context-limits/Stuffing every document into a Claude or ChatGPT project knowledge base eventually hits context limits — answers degrade first, uploads fail later. Symptoms, causes, and moving knowledge out of the context window.Mon, 17 Aug 2026 00:00:00 GMThttps://langhuan.dev/en/blog/pgvector-chinese-hybrid-search/https://langhuan.dev/en/blog/pgvector-chinese-hybrid-search/Store vectors in pgvector and tokenize Chinese with zhparser for full-text search, then fuse with RRF on a single SQL stack without a separate vector database.Thu, 13 Aug 2026 00:00:00 GMThttps://langhuan.dev/en/blog/hybrid-search-explained/https://langhuan.dev/en/blog/hybrid-search-explained/Why pure vector search misses exact keywords, what full-text search adds, how RRF deterministically fuses two ranked lists, and what it costs.Wed, 12 Aug 2026 00:00:00 GMThttps://langhuan.dev/en/blog/mcp-over-http-vs-stdio/https://langhuan.dev/en/blog/mcp-over-http-vs-stdio/Comparing stdio, SSE, and streamable HTTP transports for MCP: when to use each, and the deployment and security trade-offs.Tue, 11 Aug 2026 00:00:00 GMThttps://langhuan.dev/en/blog/what-is-mcp/https://langhuan.dev/en/blog/what-is-mcp/What the Model Context Protocol solves, what servers, clients, and tools are, and how a knowledge base becomes an MCP-callable service.Mon, 10 Aug 2026 00:00:00 GMThttps://langhuan.dev/en/blog/hello-langhuan/https://langhuan.dev/en/blog/hello-langhuan/When building several enterprise agents, I found the thing most often rebuilt isn't model capability — it's the ingestion, access control, retrieval, and evidence chain for knowledge.Wed, 05 Aug 2026 00:00:00 GMT