← Back to Labs

RAG & HNSW Vector Retrieval

Step through document chunking, dense vector embeddings, hierarchical small world graph traversal, and hybrid Reciprocal Rank Fusion

system_spec.mdChunk #1 Window (512t)Overlap Zone (128t)AST ParserOverlap: 128 tokensWindow: 512 tokensChunk #1 [0..512]Ingestion & AST semantic parsing...Overlap: HeadChunk #2 [384..896]Vector Indexing & HNSW layers...Overlap: 128 TokensChunk #3 [768..1280]Hybrid RRF & Cross-Encoder...Overlap: Tail
STEP 1 OF 6

Document Parsing & Overlapping Chunking

Large raw documents must be broken down into smaller, manageable text chunks before vector search. Using an overlapping sliding window ensures key context across sentence or block boundaries isn't lost during retrieval.

Arrow keys to navigate · R to reset

Tap dots to jump to any step

Read the full article →Take the quiz →