https://abdulrehmanbaber.dev/writing Notes from the lab — evals and anti-hallucination, observability, agents and MCP, and the AI-search visibility (GEO) niche. en-us a.r.b.plato@gmail.com (Abdul Rehman Baber) a.r.b.plato@gmail.com (Abdul Rehman Baber) Tue, 24 Mar 2026 00:00:00 GMT https://abdulrehmanbaber.dev/writing/one-mcp-server-four-clients https://abdulrehmanbaber.dev/writing/one-mcp-server-four-clients Agents & MCP Tue, 24 Mar 2026 00:00:00 GMT Pointing Cursor, a raw SDK client, LangChain, and LangGraph at one MCP server to test "write once, call from any agent" — and why transport choice, structured I/O, and not being an SSRF proxy are the real work. https://abdulrehmanbaber.dev/writing/teaching-rag-to-say-i-dont-know https://abdulrehmanbaber.dev/writing/teaching-rag-to-say-i-dont-know RAG, Evals & Reliability Tue, 10 Mar 2026 00:00:00 GMT Two gates — a model-free score floor and an evidence-required reranker that must name what is missing — plus a negative control in the eval set, so a retrieval system can honestly return "nothing here fits." https://abdulrehmanbaber.dev/writing/the-bug-that-made-our-alerts-lie https://abdulrehmanbaber.dev/writing/the-bug-that-made-our-alerts-lie Production war stories Tue, 24 Feb 2026 00:00:00 GMT A severity-1 observability blind spot: a log-shipping prefix broke level extraction, so "no errors" really meant "we stopped being able to see errors." Why a zero is a question, not an answer. https://abdulrehmanbaber.dev/writing/five-visibility-states https://abdulrehmanbaber.dev/writing/five-visibility-states AI Visibility / GEO Tue, 10 Feb 2026 00:00:00 GMT A five-state model for AI-answer visibility — full, mention-only, citation-only, third-party, invisible — and why the mention–citation gap is the cheapest thing to measure and the most expensive to ignore.