# The 8 Production RAG Architectures & Level 7 Ingestion

Category: Retrieval & Pipelines
Stack: LightRAG, GraphRAG, BM25, Multi-Modal Embeddings

## Overview
Systematic breakdown of Naive, Multimodal, HyDE, Corrective (CRAG), Graph, Hybrid, Adaptive, and Agentic RAG tiers, and when to stop at Level 4.

## The Problem
Defaulting to agentic ReAct loops for every retrieval query causes severe cost inflation (500 to 5,000+ tokens) and 15-30s latencies where simple vector or hybrid search suffices.

## Production Rule
Match retrieval architecture to query complexity. Reserve Agentic RAG for multi-step reasoning across decoupled systems. Use data normalization and content hashes for idempotent processing.

## Code / Specification
File: `rag_decision_matrix.md`

```system design
| Architecture | Core Mechanism | Use Case |
|---|---|---|
| Naive RAG | Query -> embed -> vector DB -> LLM | Single-turn Q&A, clean homogeneous docs |
| Multimodal RAG | Shared embedding space text/images | Visual charts, scanned PDFs |
| HyDE | LLM generates hypothetical doc first | Vocabulary mismatch between user & corpus |
| Corrective (CRAG) | Evaluates retrieved docs; web fallback | High precision; internal data has gaps |
| Graph RAG | Entity relationship graph | Entity connections (legal, corporate structures) |
| Hybrid RAG | Vector + keyword (BM25) | Exact terminology AND semantic meaning |
| Adaptive RAG | Query analyzer routes retrieval mode | Diverse user question patterns |
| Agentic RAG | Multi-agent ReAct cycles with memory | Multi-step cross-system investigations |
```