RAG Answer Generation

AI ↔ AI

Embed query, retrieve context, generate answer, validate grounding.

5 nodes · 5 edgesai

agentdbapi

Visual

Embed User Queryapi

Convert natural language query to vector embedding via embedding model.

↓sequential→ Vector Search

Vector Searchdb

Search vector store for top-k relevant document chunks.

↓sequential→ Generate Answer

Generate Answeragent

Produce answer grounded in retrieved context passages.

↓sequential→ Hallucination Check

Hallucination Checkagent

Verify every claim in the answer traces back to a retrieved chunk.

↓conditional→ Return Response

↓fallback→ Generate Answer

Return Responseapi

Deliver validated answer to user with source citations.

Open in Editor

uc-rag-pipeline.osop.yaml

osop_version: "1.0"
id: "rag-pipeline"
name: "RAG Answer Generation"
description: "Embed query, retrieve context, generate answer, validate grounding."

nodes:
  - id: "embed_query"
    type: "api"
    name: "Embed User Query"
    description: "Convert natural language query to vector embedding via embedding model."

  - id: "retrieve"
    type: "db"
    name: "Vector Search"
    description: "Search vector store for top-k relevant document chunks."
    timeout_sec: 5

  - id: "generate"
    type: "agent"
    subtype: "llm"
    name: "Generate Answer"
    description: "Produce answer grounded in retrieved context passages."
    security:
      risk_level: "low"

  - id: "validate"
    type: "agent"
    subtype: "llm"
    name: "Hallucination Check"
    description: "Verify every claim in the answer traces back to a retrieved chunk."

  - id: "deliver"
    type: "api"
    name: "Return Response"
    description: "Deliver validated answer to user with source citations."

edges:
  - from: "embed_query"
    to: "retrieve"
    mode: "sequential"
  - from: "retrieve"
    to: "generate"
    mode: "sequential"
  - from: "generate"
    to: "validate"
    mode: "sequential"
  - from: "validate"
    to: "deliver"
    mode: "conditional"
    when: "validation.grounded == true"
  - from: "validate"
    to: "generate"
    mode: "fallback"
    label: "Regenerate with stricter prompt"

←

Multi-Agent Research

Agent-to-Agent Handoff

→