busca-hibrida-local-first

What problem does it solve? Fictional writing projects contain invented names, places, and magic terms that pure vector search fails to retrieve because embedding models lack these out-of-vocabulary words. This Skill combines dense vector search with lexical BM25 matching so authors can reliably find lore, notes, and chapters offline on their own device. ## Core Features & Use Cases - Hybrid Retrieval with RRF: Runs SQLite-VSS vector search and BM25 lexical search in parallel, then fuses rankings with Reciprocal Rank Fusion (k=60) for balanced semantic and exact-term results. - Local-First Indexing: Chunks text into 384-token segments with 10% overlap, generates embeddings with a local multilingual-e5-base model, and indexes vectors in SQLite-VSS using HNSW (M=16, ef_construction=200). - Use Case: A novelist searches "Where was Morgan's staff broken?" and the engine returns the exact chapter mentioning the staff, ranking exact lore terms above merely similar concepts like "scepter". ## Quick Start Ask the assistant to index my novel chapters and search for where a specific fictional item or character appears using hybrid local search.