Domain Retrieval Router & Hybrid Vector Search Workbench

Decomposition & Routing

Multi-Intent Query Presets

Query Vector Decomposition Input

Routing Similarity Threshold 0.65

Vectors below threshold route to general fallback pool

Specialized Domain Index Weights

Milkman / Dairy

Butcher / Meats

Greengrocer

Coalman / Fuel

Pop Lorry / Drinks

Retrieval Telemetry & Benchmarks

Token Context Efficiency 78.4% Reduction vs Monolithic Dense Prompt

Sub-Query Precision 0.96 Specialized domain precision match

Context Window Load 480 tokens Monolithic baseline: 2,450 tokens

Noise & Hallucination Risk 12.1% Irrelevant document chunk overhead

Vector Sub-Intent Breakdown

Sub-Intent Clause	Routed Index	Similarity	Status

Domain Router Architecture

Query is split into sub-intents matched against localized indices (Milkman, Greengrocer). Specialized retrieval avoids bloat by inserting only relevant domain chunks into LLM prompt context.

Decomposition & Routing

Retrieval Telemetry & Benchmarks

RAG Pipeline & Domain Routing Package