RAG Vector Retrieval & Context Grounding Simulator

2D Vector Space Projection (Embedding Space) Click node to inspect chunk details

User Query Vector

Retrived Context (Top-K)

Unretrieved Chunks

Grounded Context Window (Prompt Assembly) 114 / 2048 Tokens

RAG Telemetry & Grounding

Retrieved Chunks 2

Top Similarity Score 0.89

Context Tokens 114

Hallucination Risk Low (12%)

Selected Chunk Inspector

Chunk #doc-2-0 Score: 0.890

Document: Context Window Packing

"Retrieval Augmented Generation injects relevant text passages directly into the prompt context window before inference."

Cosine Similarity: 0.89

Dot Product: 1.42

Euclidean Distance: 0.47

RAG Physics Guide

Cosine Similarity: Measures the angular vector alignment regardless of vector magnitude. Key metric for semantic similarity in text embeddings.

Hallucination Guardrail: By filtering chunks below threshold 0.72, out-of-domain query hallucinations drop exponentially.