WAVGEN Pipeline Embeddings Chunking Retrieval Reranking Builders Glossary

RAG SYSTEMS

RETRIEVAL-AUGMENTED GENERATION / VECTOR SEARCH / SEMANTIC MEMORY / EMBEDDINGS

§ 02 — RAG PIPELINE

Retrieval-Augmented Generation Architecture

RAG grounds LLM generation in verified external knowledge by retrieving semantically relevant documents at query time and injecting them into the prompt context — reducing hallucination and enabling real-time knowledge access.

Documents

PDFs, web pages, code, Markdown, databases, APIs

Chunking

Split into semantically coherent units with overlap strategy

Embedding

Convert chunks → dense vectors via embedding model

Index

Store vectors + metadata in vector database (HNSW / IVF)

Retrieve

Embed query → ANN search → top-K chunks

Augment

Inject retrieved context into LLM prompt template

Generate

LLM produces grounded response from augmented context

LIVE RAG QUERY FLOW VISUALIZER

§ 03 — EMBEDDING SPACE

Semantic Embedding Explorer

Embedding models compress text into dense high-dimensional vectors where semantic similarity corresponds to geometric proximity. The resulting space encodes meaning: synonyms cluster, antonyms diverge, analogical relationships form parallelogram structures.

2D PROJECTED EMBEDDING SPACE — MOVE CURSOR TO QUERY

COSINE SIMILARITY HEATMAP

EMBEDDING DIMENSION DISTRIBUTION

§ 04 — CHUNKING STRATEGIES

Document Chunking Architecture

Chunking strategy fundamentally determines retrieval precision. Fixed-size, sentence-aware, recursive, semantic, and proposition-level chunkers each produce different recall/precision trade-offs depending on document structure and query type.

CHUNKING STRATEGY COMPARISON — DRAG SLIDER

CHUNK PARAMETERS

Chunk Size (tokens)512

Overlap (tokens)64

Min Chunk Size100

Separator LevelPARA

INDEXING CONFIG

Embedding Model Dim1536

Index TypeHNSW

HNSW M16

ef_construction200

§ 05 — SEMANTIC RETRIEVAL

Vector Retrieval Simulation

At query time, the input is embedded into the same vector space as indexed chunks. Approximate nearest neighbor search returns the top-K most semantically similar chunks — the geometry of the embedding space determines recall quality.

ANN RETRIEVAL SIMULATION — CURSOR IS YOUR QUERY VECTOR

HNSW INDEX LAYER TRAVERSAL

RETRIEVED CHUNK RELEVANCE

§ 06 — RERANKING & FUSION

Retrieval Reranking Architecture

Initial vector retrieval is fast but noisy. A cross-encoder reranker re-scores each candidate chunk against the full query with full attention — dramatically improving precision at the cost of additional latency. Hybrid BM25+vector fusion further improves recall.

RERANKING PIPELINE — BEFORE / AFTER COMPARISON

RETRIEVAL SYSTEM PERFORMANCE

Vector Search Recall@5

78%

After Reranking Precision@5

91%

BM25 Recall@5

68%

Hybrid Fusion Recall@5

88%

Reranker Latency Added

+35ms avg

§ 07 — CONTEXTUAL MEMORY

Contextual Memory Systems

Advanced RAG architectures maintain multi-tier memory: dense vector stores for semantic search, sparse BM25 indexes for keyword precision, knowledge graphs for structured reasoning, and conversation history for session continuity.

MULTI-TIER MEMORY ARCHITECTURE

§ 08 — BUILDERS

RAG Ecosystem Builders

The researchers and engineers shaping the retrieval-augmented generation landscape.

§ 09 — GLOSSARY

RAG Systems Lexicon

Core vocabulary for retrieval-augmented generation architecture and implementation.