Setting Up a Demo with RAG Engine on Gemini Enterprise Agent Platform

This document describes the steps required to set up a simple RAG demo using Vertex AI and Gemini Enterprise Agent Platform. The goal is to validate the end-to-end document ingestion and retrieval workflow before integrating it into backend services.

For a more detailed overview of RAG Engine concepts, supported architectures, and implementation options, refer to the official Google documentation: RAG Engine on Gemini Enterprise Agent Platform Overview .

Prerequisites

• Google Cloud SDK installed
• Access to a Google Cloud project
• Python 3.10 or later
• Vertex AI permissions enabled

---

1. Authenticate with Google Cloud

Login using Application Default Credentials:

gcloud auth application-default login

Verify the active project:

gcloud config get-value project

The command should return your target project ID.

---

2. Enable Required Services

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

Verify that Vertex AI is enabled:

gcloud services list | grep aiplatform

---

3. Create a Storage Bucket (UI)

Open Google Cloud Storage

Create a bucket named:

Rag-demo

Select an appropriate region for your environment.

---

4. Upload Source Documents (UI)

Open the Rag-demo bucket and upload the documents that will be used for retrieval. Supported formats include PDF, TXT, JSON, and other text-based files.

Optional verification:

gcloud storage ls gs://Rag-demo/

---

5. Create a Python Environment

python3 -m venv .venv
source .venv/bin/activate

Install dependencies:

pip install google-cloud-aiplatform

---

6. Open Gemini Enterprise Agent Platform (UI)

Open Gemini Enterprise Agent Platform

If prompted, enable the required APIs before proceeding.

---

7. Create a RAG Corpus

Create create_rag.py:

import vertexai
from vertexai.preview import rag
from vertexai.preview.rag.utils.resources import (
    RagVectorDbConfig,
    RagManagedDb,
)

vertexai.init(
    project="YOUR_PROJECT_ID",
    location="YOUR_REGION",
)

corpus = rag.create_corpus(
    display_name="demo-rag",
    backend_config=RagVectorDbConfig(
        vector_db=RagManagedDb()
    ),
)

print(corpus.name)

Run:

python3 create_rag.py

Save the generated corpus ID. It will be required in the next steps.

---

8. Import Documents into the Corpus

Create import_pdf.py:

import vertexai
from vertexai.preview import rag

vertexai.init(
    project="YOUR_PROJECT_ID",
    location="YOUR_REGION"
)

response = rag.import_files(
    corpus_name="YOUR_CORPUS_ID",
    paths=[
        "gs://Rag-demo/YOUR_DOCUMENT.pdf"
    ]
)

print(response)

Run:

python3 import_pdf.py

Expected output:

imported_rag_files_count: 1

---

9. Run a Test Query

Create query.py:

import vertexai
from vertexai.preview import rag

vertexai.init(
    project="YOUR_PROJECT_ID",
    location="YOUR_REGION"
)

response = rag.retrieval_query(
    rag_resources=[
        rag.RagResource(
            rag_corpus="YOUR_CORPUS_ID"
        )
    ],
    text="Summarize this document"
)

print(response)

---

Notes

• JSON and plain text documents generally provide better retrieval quality than large spreadsheets or multi-column PDFs.
• If your source data is stored in Excel files, converting the data into structured JSON before importing into RAG is strongly recommended.
• Recommended workflow:

Excel (.xlsx)
↓
Transform to JSON
↓
Generate clean text documents
↓
Import into RAG Engine

Structured JSON enables better chunking, more accurate embeddings, and significantly improves retrieval quality compared to importing spreadsheets directly.

---

Next Steps

• Integrate retrieval logic into backend services.
• Create a dedicated /rag/ask endpoint.
• Connect frontend applications to the RAG service.
• Introduce preprocessing pipelines for Excel-to-JSON conversion.