SJ88 AI Knowledge Hub — Field notes https://sj88ai.com/#blog Reality-first posts on AI workflows, prompting, RAG, and what actually works in production. en-th Five prompt patterns I use every day (and the one I stopped using) https://sj88ai.com/#blog https://sj88ai.com/#blog/prompt-patterns-2026 Tue, 30 Jun 2026 09:00:00 +0700 Prompting A year of daily LLM work distilled into the patterns that actually move the needle — and the once-popular pattern I now actively avoid. Why your RAG app hallucinates — and the chunking fix that worked for me https://sj88ai.com/#blog https://sj88ai.com/#blog/rag-chunking-fix Mon, 15 Jun 2026 14:30:00 +0700 RAG A 6-month debugging diary: from 30% to 92% answer accuracy by changing one thing about how I split documents. Local LLMs in 2026: my actual hardware budget vs. what I expected https://sj88ai.com/#blog https://sj88ai.com/#blog/local-llm-budget Wed, 28 May 2026 18:00:00 +0700 Local AI I built the same RAG app on three local stacks (Ollama, LM Studio, llama.cpp) and ran it for 30 days. Here's what I learned about hardware, latency, and what breaks.