Return_to_Blueprints
Architecture/Phase_07/22 min build

Multi-Index RAG with Fallback Cascade

A rigorous, 9-node RAG pipeline. It queries a primary, highly-specific vector index first. If confidence is low, it falls back to a broad archive index, reranking results to ensure hallucination-free legal synthesis.

Execution_Steps

  1. 01

    Primary Retrieval Vector

    Link an `input` node to an `embedding` node, then to the primary `vector_store` ("Active Contracts"). Connect this to a `router` acting as a Confidence Gate.

  2. 02

    Build the Primary Path

    If the Confidence Gate passes, route traffic to a `reranker` to distill the Top 10 chunks down to the Top 3 most relevant. Route this directly to your `llm_call` synthesizer.

  3. 03

    Build the Fallback Cascade

    Connect the Miss path of the router to a secondary `vector_store` ("Archive Precedents"). Connect this to a separate `reranker` configured for Top-K: 5.

  4. 04

    Converge and Generate

    Connect the fallback reranker to the exact same `llm_call` synthesizer node used in the primary path. Finally, terminate the DAG with an `output` node to serve the legal brief.

Expected_Metrics

P50_LATENCY:< 5200ms
COST_SAVING:10.5%
SLA_LIMIT:7000ms

Ready to verify?

Open the canvas and simulate these parameters in real-time.

Node_Architecture

inputLegal QueryText Mode
embeddingContract Embeddertext-embedding-3-large
vector_storeActive ContractsTop-K: 10
routerConfidence GateLogic: Score > 0.85
vector_storeArchive PrecedentsFallback / Top-K: 20
rerankerPrimary RerankerCohere rerank-v3.5 / Top-K: 3
rerankerFallback RerankerCohere rerank-v3.5 / Top-K: 5
llm_callLegal Counsel LLMgpt-4o
outputFinal BriefStandard

Recursive_Read_Next

Architecture

The AI Essay Detector

Data Pipeline

Enterprise Support RAG Pipeline