Appendix - RAG (Retrieval-Augmented Generation)

What We Learned

Parse, chunk and index documents with metadata, versions and access controls.
Combine lexical and vector retrieval; rerank candidates before assembling answer context.
Ground answers in evidence with checked citations and abstention when evidence is insufficient.
Evaluate retrieval and generation separately to locate failures.
Choose advanced, agentic, long-context or managed patterns from task needs; maintain freshness and permissions in production.

Term	Short meaning
RAG / CAG	Retrieval-Augmented / Cache-Augmented Generation: retrieved evidence / cached knowledge context.
Embedding / vector search	Numeric representation of meaning / search by vector similarity.
ANN / HNSW	Approximate Nearest Neighbor / Hierarchical Navigable Small World: fast approximate search / a graph index.
Chunk / overlap	Retrieval-sized text unit / content shared between neighboring units.
BM25	Lexical ranking using term frequency, document frequency and length normalization.
Hybrid retrieval / RRF	Combines lexical and semantic search / Reciprocal Rank Fusion combines ranked lists.
Reranker	Rescores retrieved candidates for query relevance.
Recall@k / MRR	Relevant-item coverage in top k / Mean Reciprocal Rank of the first relevant result.
nDCG	Normalized Discounted Cumulative Gain: ranking quality with relevance and position weighting.
Grounding / faithfulness	Tying claims to evidence / consistency of the answer with that evidence.
Citation / abstention	Traceable source reference / declining an unsupported answer.
GraphRAG / agentic RAG	Graph-based evidence retrieval / an agent-controlled retrieval loop.
ACL / CDC	Access Control List / Change Data Capture: permissions / source change tracking.
Freshness / lineage	How current indexed data is / its source and transformation history.