A · PREPARE
Chunking
Splits large content into retrievable pieces.
Embeddings
Represent meaning as vectors.
Vector Database
Stores and indexes vectors efficiently.
08 · Embeddings, Search & RAG
How AI Finds External Information Before Answering
This module explains how documents can be prepared, searched, retrieved, and added to a model’s context so answers can use relevant external information.
Understand the full path from source documents to retrieved context.
The Big Picture · two information flows converge
Prepare knowledge, then find what matters
KNOWLEDGE PREPARATION
DOCUMENTSSource material
↓
CHUNKINGSplit large content into useful pieces
↓
EMBEDDINGSRepresent chunks as vectors
↓
VECTOR DATABASE / INDEXStore and organize searchable representations
QUERY TIME
USER QUESTIONA request for information
↓
QUERY REPRESENTATIONRepresent the request for search
↓
SEARCHSemantic Search / Vector Search
↓
RETRIEVAL + RERANKINGReturn and order candidate source content
RELEVANT CONTEXT
↓ RAG · add retrieved context to generation
↓ MODEL → GROUNDED ANSWER
Vector databases are one common implementation, not a requirement for every retrieval system.
Knowledge Map
Prepare → Find → Use
B · FIND
Semantic Search
Searches by meaning, not only exact words.
Vector Search
Finds mathematically similar vectors.
Retrieval
The broader process of finding useful source information.
Reranking
Reorders candidates by relevance.
C · USE
RAG
Adds retrieved information to model context before generation.
Grounding
Connects an answer to supplied external evidence.
Key Distinctions
Connected concepts, different layers
EmbeddingRepresentation≠Vector DatabaseStorage and indexing system
Semantic SearchSearch by meaning≠Vector SearchOne technical method
Vector SearchSearch method≠RetrievalBroader information-finding process
RetrievalFinds information≠RAGUses it during generation
RAGRuntime context≠Fine-tuningAdditional training
GroundingTies output to evidence≠Guaranteed correctnessStill requires evaluation
One Example
Answering a leave-policy question
KNOWLEDGE BASE
Employee handbook → chunk leave policy → index content
USER REQUEST
“How many annual leave days do employees receive?” → search / retrieval → relevant leave-policy chunk → add to LLM context → answer
The answer can reference the retrieved policy.
Recommended Learning Path
Learn the retrieval system in order
This is a learning path, not a mandatory technical stack.
01Chunking
Divide documents.
02Embeddings
Represent meaning.
03Vector Database
One storage option.
04Search
Semantic and vector search.
05Retrieval
Select useful content.
06Reranking
Improve ordering.
07RAG
Combine retrieval and generation.
08Grounding
Connect answers to evidence.
Remember the pipeline
Retrieval systems prepare and search external knowledge, while RAG supplies the most relevant information to a model before it generates an answer.