0-4-6-understand-embedding-model

name	0.4.6-understand-embedding_model
description	[0.4.6] Neural model converting text into dense vector representations optimized for retrieval similarity tasks

understand-embedding_model

CALL NUMBER: retrieval_augmented_generation_architecture_patt.embedding_model DEFINITION: Neural model converting text into dense vector representations optimized for retrieval similarity tasks

Invoke this skill to understand embedding_model down to its primitives. The RELATIVE ROOT below is the least-fixed-point closure of everything it bundles from — the full import cone, grouped by the lib each prim comes from. Projected from a prover-typed KB (MAP/SWI-Prolog consistency gate): every reference below resolves.

THE RELATIVE ROOT (the import cone, by lib)

from retrieval_augmented_generation_architecture_patt

• bi_encoder_retrieval (d1): Embedding approach encoding query and documents independently into fixed vectors enabling fast similarity computation
• embedding_dimensionality (d1): Vector size of text representations — trade-off between expressiveness, storage cost, and retrieval speed
• few_shot_retrieval (d1): Retrieval improved through in-context examples demonstrating relevance judgments without gradient updates
• zero_shot_retrieval (d1): Retrieval on unseen domains without training using zero-shot capable embedding models
• dense_retrieval (d2): Neural retrieval using learned embedding models to encode queries and documents into dense vectors for similarity search
• approximate_nearest_neighbor (d3): Vector indexing technique trading exactness for speed: HNSW, IVF, or LSH structures enabling sub-linear retrieval at scale
• cross_encoder_reranking (d3): Two-pass retrieval: initial ANN retrieval followed by full cross-encoder scoring of candidate-document pairs for refined ranking
• dense_passage_retriever (d3): Bi-encoder model trained on query-passage pairs to produce jointly learned dense embeddings for retrieval
• hybrid_rag (d3): RAG combining dense vector retrieval with sparse lexical retrieval (BM25) to leverage both semantic similarity and exact keyword matching
• hybrid_search (d3): Retrieval combining multiple search paradigms (dense/sparse, vector/keyword) with score normalization or late fusion
• colbert_style_retrieval (d4): Late interaction retrieval: compute token-level similarity then aggregate for fine-grained relevance without full cross-encoding
• retrieval_fusion (d4): Combining results from multiple retrieval methods through Reciprocal Rank Fusion, Score Normalization, or learned weights
• reciprocal_rank_fusion (d5): Fusing ranked retrieval lists from multiple systems by reciprocal rank weighting to produce unified ranking

CONSUMERS (what needs this)

domain_adaptation_rag

---

Projected from the retrieval augmented generation architecture patterns KB (207 concepts / 225 relations) — consistency-typed by MAP; the facet list after the colon IS the cross-lib dependency web.

(leaf — this is an actual skill.)