Skip to content

README.md

Latest commit

 

History

History
194 lines (144 loc) · 5.92 KB

README.md

File metadata and controls

194 lines (144 loc) · 5.92 KB
Raw

08. Vertex AI RAG Engine

Retrieval-Augmented Generation (RAG) with Vertex AI - Build agents that answer questions using your own documents.

Blog Post: https://arjunprabhulal.com/adk-builtin-tools-rag/

Table of Contents

  1. What is RAG?
  2. Prerequisites
  3. Google Cloud Setup
  4. Create a RAG Corpus
  5. Setup Steps
  6. Running the Agent
  7. Project Structure
  8. Next Steps

What is RAG?

RAG (Retrieval-Augmented Generation) enhances LLM responses by:

  1. Retrieving relevant documents from your knowledge base
  2. Augmenting the prompt with retrieved context
  3. Generating accurate answers grounded in your data
┌─────────────────────────────────────────────────────────────┐
│                      USER QUERY                              │
│              "What is our refund policy?"                    │
└─────────────────────────────────────────────────────────────┘
                            │
                            ▼
┌─────────────────────────────────────────────────────────────┐
│                   RAG RETRIEVAL                              │
│                                                              │
│   Query → Vector Search → Top-K Documents                    │
│                                                              │
│   Retrieved: "policy.pdf", "faq.md", "terms.txt"            │
└─────────────────────────────────────────────────────────────┘
                            │
                            ▼
┌─────────────────────────────────────────────────────────────┐
│                   LLM GENERATION                             │
│                                                              │
│   Context: [Retrieved documents]                             │
│   Query: "What is our refund policy?"                        │
│                                                              │
│   Response: "Based on our policy documents, refunds are..."  │
└─────────────────────────────────────────────────────────────┘

Prerequisites

  • Python 3.10+
  • Google Cloud Project with billing enabled
  • gcloud CLI installed and authenticated
  • Documents to upload (PDF, TXT, HTML, etc.)

Google Cloud Setup

1. Enable Required APIs

gcloud services enable aiplatform.googleapis.com
gcloud services enable discoveryengine.googleapis.com

2. Grant IAM Roles

# Replace with your project ID and email
PROJECT_ID="your-project-id"
USER_EMAIL="your-email@example.com"

# Vertex AI User - for using AI Platform
gcloud projects add-iam-policy-binding $PROJECT_ID \
  --member="user:$USER_EMAIL" \
  --role="roles/aiplatform.user"

# Discovery Engine Editor - for RAG operations
gcloud projects add-iam-policy-binding $PROJECT_ID \
  --member="user:$USER_EMAIL" \
  --role="roles/discoveryengine.editor"

3. Authenticate

gcloud auth application-default login

Create a RAG Corpus

Option A: Using Google Cloud Console

  1. Go to Vertex AI Console
  2. Navigate to RAG Engine → Corpora
  3. Click Create Corpus
  4. Upload your documents
  5. Copy the corpus resource name

Option B: Using Python SDK

from vertexai.preview import rag

# Create corpus
corpus = rag.create_corpus(display_name="my-knowledge-base")

# Upload documents
rag.upload_file(
    corpus_name=corpus.name,
    path="path/to/document.pdf",
    display_name="document.pdf"
)

print(f"Corpus: {corpus.name}")

Setup Steps

  1. Navigate to this module:
cd 08-vertex-ai-rag
  1. Create and activate a virtual environment:
python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate
  1. Install dependencies:
pip install -r ../requirements.txt
  1. Configure environment variables in rag_agent/.env:
GOOGLE_GENAI_USE_VERTEXAI=1
GOOGLE_CLOUD_PROJECT=your-project-id
GOOGLE_CLOUD_LOCATION=us-central1
RAG_CORPUS=projects/your-project-id/locations/us-central1/ragCorpora/your-corpus-id

Replace your-project-id and your-corpus-id with your actual values.

Running the Agent

Using ADK Web

adk web

Open http://127.0.0.1:8000 and select rag_agent.

Test Queries:

  • "What documents do you have access to?"
  • "Summarize the main topics in my knowledge base"
  • "Find information about [your document topic]"

Using ADK CLI

adk run rag_agent

Project Structure

08-vertex-ai-rag/
├── README.md
└── rag_agent/
    ├── __init__.py
    ├── agent.py          # RAG agent with VertexAiRagRetrieval tool
    └── .env              # Vertex AI configuration

Key Configuration Options

Parameter Description Default
similarity_top_k Number of documents to retrieve 10
vector_distance_threshold Similarity threshold (0-1) 0.6
rag_resources List of corpus resources Required

Next Steps

Continue to 09. Vertex AI Search