01
Agentic RAG
Definition
A retrieval workflow where an agent plans searches, uses tools, evaluates evidence, and iterates before answering.
Back to Agentic AI Glossary
Agentic AI Glossary
RAG & Vector Retrieval Terms
RAG & Vector Retrieval Terms terms and explanations from the Agentic AI Glossary.
66 terms in this chapter
02
Answer Relevance
Definition
How well an answer addresses the user's question or task.
03
Approximate Nearest Neighbor
Definition
A fast search method that finds near-best vector matches efficiently at scale.
04
BM25
Definition
A classic keyword ranking algorithm used in sparse retrieval and hybrid search.
05
Chunk Overlap
Definition
Repeated text shared between adjacent chunks to preserve context across boundaries.
06
Chunk Size
Definition
The amount of text placed into each retrieval unit during document processing.
07
Chunking
Definition
Splitting documents into smaller passages so retrieval can return useful, focused context.
08
Citation Generation
Definition
Producing references or source links that show where an answer was grounded.
09
Context Compression
Definition
Reducing retrieved content to the most relevant facts before sending it to the model.
10
Context Precision
Definition
The share of retrieved context that is actually useful for answering the question.
11
Context Recall
Definition
How much of the needed evidence the retrieval system successfully found.
12
Context Relevance
Definition
How relevant retrieved context is to the query and final answer.
13
Contextual Retrieval
Definition
Retrieval that uses surrounding context, metadata, or rewritten queries to improve relevance.
14
Corrective RAG
Definition
A RAG pattern that detects weak retrieval or unsupported answers and corrects the retrieval or response path.
15
Cosine Similarity
Definition
A measure of vector similarity based on the angle between vectors.
16
Cross-Encoder Reranker
Definition
A reranker that jointly reads the query and candidate passage to produce a relevance score.
17
Document Ingestion
Definition
The process of loading, parsing, cleaning, and preparing source documents for retrieval.
18
Document Parsing
Definition
Extracting text, tables, metadata, and structure from source files or webpages.
19
Dot Product Similarity
Definition
A similarity score calculated by multiplying and summing vector components.
20
Embedding
Definition
A vector representation of data used for semantic search and similarity matching.
21
Faithfulness
Definition
The degree to which an answer is supported by the provided context.
22
GraphRAG
Definition
A RAG approach that uses graph relationships among entities, documents, and facts to improve reasoning.
23
Grounded Response
Definition
An answer supported by retrieved evidence, source-of-truth data, or citations.
24
Hallucination Detection
Definition
Methods for identifying unsupported or fabricated model claims.
25
Hybrid Search
Definition
Combining keyword and vector retrieval to improve recall and relevance.
26
Keyword Search
Definition
Search based on exact words, phrases, or lexical matching.
27
Long-Context RAG
Definition
A retrieval approach that combines large context windows with selected external evidence.
28
Metadata Filtering
Definition
Restricting retrieval by document attributes such as source, date, owner, product, or permission.
29
Multi-Query Retrieval
Definition
Retrieving with several query variations to improve coverage of relevant context.
30
Query Expansion
Definition
Adding related terms or variants to a query to improve retrieval coverage.
31
Query Rewriting
Definition
Transforming a user question into a better search query or multiple focused queries.
32
RAG Pipeline
Definition
The end-to-end flow for ingestion, chunking, embedding, retrieval, reranking, and answer generation.
33
Reranking
Definition
Reordering retrieved results using a stronger relevance model after initial search.
34
Retrieval-Augmented Generation
Definition
A pattern that retrieves relevant external knowledge before generation so responses can be grounded in sources.
35
Retrieval Evaluation
Definition
Measuring whether the retrieval system returns useful and complete context.
36
Semantic Search
Definition
Search based on meaning rather than exact keyword matching.
37
Similarity Search
Definition
Finding items closest to a query representation in a vector or feature space.
38
Vector
Definition
A numeric representation of meaning, features, or model state.
39
Vector Database
Definition
A database optimized for storing embeddings and retrieving semantically similar content.
40
ANN Index
Definition
An approximate-nearest-neighbor index that finds similar vectors quickly without comparing every vector exactly.
41
Chroma
Definition
An open-source vector database commonly used for local RAG prototypes and lightweight embedding search.
42
Dense Retrieval
Definition
Dense Retrieval is a retrieval method focused on dense. It finds useful information before generation, tool use, grounding, or answer verification.
43
Elasticsearch
Definition
A distributed search engine used for keyword search, logs, analytics, and hybrid retrieval systems.
44
Embedding Drift
Definition
The term Embedding Drift means an embedding drift concept used in vector database and search terms for practical AI engineering work.
45
Embedding Space
Definition
The term Embedding Space means an embedding space concept used in vector database and search terms for practical AI engineering work.
46
FAISS
Definition
A library from Meta for efficient vector similarity search and clustering.
47
HNSW
Definition
Hierarchical Navigable Small World, a graph-based index commonly used for fast vector search.
48
Hybrid Retrieval
Definition
Hybrid Retrieval is a retrieval method focused on hybrid. It finds useful information before generation, tool use, grounding, or answer verification.
49
IVF
Definition
Inverted file indexing, a vector search method that partitions embeddings into clusters for faster lookup.
50
Metadata Store
Definition
A storage system for metadata data that an AI application can save, query, or retrieve during execution.
51
Milvus
Definition
An open-source vector database for storing, indexing, and searching large-scale embedding collections.
52
MMR
Definition
Maximal Marginal Relevance, a retrieval method that balances relevance with diversity in selected results.
53
OpenSearch
Definition
An open-source search and analytics engine used for logs, keyword search, and hybrid retrieval.
54
pgvector
Definition
A PostgreSQL extension that stores embeddings and supports vector similarity search inside Postgres.
55
Pinecone
Definition
A managed vector database service used to store and search embeddings at scale.
56
PQ
Definition
Product quantization, a compression method that reduces vector storage size for faster large-scale search.
57
Precision@k
Definition
The term Precision@k means a precision at k concept used in vector database and search terms for practical AI engineering work.
58
Qdrant
Definition
A vector database focused on similarity search, filtering, and production-ready embedding retrieval.
59
Recall@k
Definition
The term Recall@k means a recall at k concept used in vector database and search terms for practical AI engineering work.
60
Semantic Cache
Definition
Semantic Cache stores reusable semantic results. It reduces repeated work, lowers cost, and improves speed when similar requests appear again.
61
Similarity Threshold
Definition
The term Similarity Threshold means a similarity threshold concept used in vector database and search terms for practical AI engineering work.
62
Sparse Retrieval
Definition
Sparse Retrieval is a retrieval method focused on sparse. It finds useful information before generation, tool use, grounding, or answer verification.
63
Top-k Retrieval
Definition
Top-k Retrieval is a retrieval method focused on top-k. It finds useful information before generation, tool use, grounding, or answer verification.
64
Vector Index
Definition
An index structure for vector that speeds up lookup, retrieval, or similarity matching.
65
Vector Store
Definition
A storage system for vector data that an AI application can save, query, or retrieve during execution.
66
Weaviate
Definition
A vector database that combines semantic search, metadata filtering, and knowledge-oriented data modeling.
Explore more chapters or test your knowledge with quizzes.