How to Build a production-ready RAG AI agent

Опубликовано 28 марта 2026, 19:00

GCP credit →goo.gle/handson-ep2-lab2
Codelab & source code → goo.gle/scholar
Try Google ADK → goo.gle/4bPEHej

In this episode, Ayo and Annie go from structured data to a fully deployed, data-aware RAG agent, and we cover a LOT of ground. Starting where they left off from last episode (BigQuery + BQML.GENERATE_TEXT), the duo now wire up the full backend for an AI agent: a vector database, an embedding pipeline, a RAG retrieval system, and a production ready Cloud Run deployment.

🛠️ *What we build:*
* Cloud SQL for PostgreSQL with pgvector for semantic search
* A containerized Apache Beam pipeline on Dataflow to batch-process text and generate Gemini embeddings
* A RAG retrieval layer that lets the agent query vectorized knowledge
* An ADK based agent that answers questions using that knowledge
* A Cloud Run deployment with proper security and scalability settings

This is hands-on, infrastructure-meets AI content. you'll leave with a real, working pattern you can adapt for your own projects.

Chapters:
0:00 - Intro
1:41 - (RAG) Retrieval Augmented Generation and chunking
4:40 - Data project overview
4:52 - Similarity search
6:40 - RAG in BigQuery
11:56 - [BQML] ML Generate in Big Query
19:46 - OLAP & OLTP
24:21 - AI in CloudSQL
28:38 - Index using HNSW
31:29 - Scaling with data pipeline
36:46 - Apache Beam
53:02 - RAG agent With CloudSQL
1:09:52 - Flight the BOSS with A2A

More resources:
AI in CloudSQL→ goo.gle/4uRlm5v
Apache Beam → goo.gle/3O6OJzY
ADK Sample → goo.gle/4rQKWVn

Watch more Hand on AI → goo.gle/HowToWithGemini
🔔 Subscribe to Google Cloud Tech → goo.gle/GoogleCloudTech

#Gemini #GoogleCloud

Speakers: Ayo Adedeji, Annie Wang
Products Mentioned: Agent Development Kit, Dataflow