Retrieval Playground — BM25 vs Embeddings vs Hybrid

Sentence-chunk a document, retrieve with from-scratch BM25, dense embeddings (all-MiniLM-L6-v2 via Transformers.js, running in this tab), and their Reciprocal Rank Fusion — then score each against the chunks you mark relevant. BM25, RRF and the metrics are hand-written below; no retrieval libraries.