If you’re building LLM apps and feel confused about when to use keyword search, embeddings, rerankers, or vector databases, this repo is for that. I built a docs-first repo on practical LLM system design patterns, covering pre-filtering, hybrid retrieval, rerankers, in-memory scoring vs vector DBs, batching, cleanup, and LLM-as-judge evaluation, with simple Python examples. From my experience, embedding quality or RAG alone is rarely the full answer. The engineering harness around the LLM us...