Brand book
Democratizing AI:
Scalable ML with
BigQuery ML & Vertex AI
CEO, AdFlex
GDE @ Cloud (since 2016)
Tu Pham
Phuong
Google Cloud Next Proprietary
Your AI needs to
be fueled by your
business data
Your business
data needs to be
unlocked by AI
Google Cloud Next Proprietary
Agenda
01 Barriers in Industry
02 Data to AI Journey
03 Built in AI/ML in BigQuery
04 BQML Demo
Sentiment Analysis
Analyzing Images
04 Questions
Google Cloud Next Proprietary
Data and AI Trends
71%
Of organizations
plan to use
databases
integrated with
gen AI
capabilities
Of organizations
are satisfied with
their legacy data
platform’s
support for AI,
indicating there
is a lot of room
for improvement
14%
Data and AI Trends Report 2024 Data and AI Trends Report 2024
Google Cloud Next Proprietary
Businesses struggle: Meeting the
needs of the AI era
84%
Time and
complexity from
data to insights is
long due to
fragmented tools
not designed to talk
to each other
business pain point
is the need to train
and upskill
employees in the
use of data and AI
#1
Data and AI Trends Report 2024 MIT Technology Review Insights
Google Cloud Next Proprietary
The Challenge: AI/ML's
Perceived Barriers
Complexity Expertise Cost
'AI/ML is too
complex for our
team.’
'We lack specialized
data science skills.'
'Infrastructure and
resources are too
expensive.'
Accessibility
'Finding the right
tools is a challenge.’
Google Cloud Next Proprietary
Unified Platform for Data to AI
Simplifying and
unifying data
analytics & ML
around BigQuery
BigQuery
Unified Platform from Data to AI
Unified experience
Single product UX
New GenAI powered
experiences
Collaborative workflows
Unified data Unified engines
Structured / unstructured
Iceberg / Delta / Hudi
GCS
Cross-cloud (AWS, Azure)
Unified governance
SQL
Spark
Python
Remote functions
Business intelligence
Google Cloud Next Proprietary
Typical ML workflow poses
many challenges
Issues:
Streaming/batch data
Data processing Export data
Manage infra?
Learn a new
language?
Data
Governance?
Deploy ML
model
Train ML model
(e.g. Python)
Where do I
host?
Multiple products & roles can lead
to unnecessary complexity & costs
BigQuery
(data warehouse)
Google Cloud Next Proprietary
BigQuery ML brings AI/ML
capabilities to your data platform
Issues:
Streaming/batch data
Data processing Export data
Manage infra?
Data
Governance?
Deploy ML model
Train ML model
(e.g. Python)
Where do I
host?
Simplify with BigQuery ML - data and
ML workloads in one place
BigQuery
(data warehouse)
Google Cloud Next Proprietary
Built-in AI/ML | BQML
All
workloads
Why BigQuery?
VS
➢ Require every ML use case to go
through more specialized systems
that require advanced skill sets
➢ Provide ML access to
more users through a
simple SQL interface
Machine Learning for all Built-in ML with
SQL
➢ Execute, iterate, and automate ML initiatives all within
BigQuery using predefined models
➢ Leverage external models developed in Tensorflow
directly from SQL
➢ Export developed models for use in Vertex AI
Google Cloud Next Proprietary
Full suite of ML platform capabilities
Feature
engineering
Inference
Train model
BigQuery
Tables or
Object tables
★ Rich built in ML
featurization functions
e.g. bucketize, encode,
normalize, resize image
★ Feature transformation
modules
★ Generate embeddings
★ Feature store
integration
★ Training managed by
BQ in BQ infra and in
VertexAI
★ Choose from 18 built
in models covering
major use cases
★ HPO support
★ Batch inference via SQL
★ Import model files (tf,
xgboost, onnx) for inference
★ Remote inference via Vertex
MaaS or user managed
endpoints
★ Streaming inference with
continuous queries
Tune, evaluate,
register models
★ Model monitoring
★ Model evaluation for
LLMs and predictive
ML models
★ Register to Vertex
Model Registry
Monitoring
Google Cloud Next Proprietary
GenAI using BigQuery ML
Vertex AI
GCP’s AI Platform
BigQuery ML
SQL interface to Vertex
AI Gemini, 3P and OSS
models
BigQuery
Cloud Data Warehouse
Model as a Service (via Vertex AI)
1P
Pro, Flash
3P Open Models
Hugging Face models deployed via Model
Garden can be used in BigQuery ML
Via User managed
Vertex AI endpoints
User Managed endpoints
Google Cloud Next Proprietary
Generative AI use cases
in BigQuery
Entity
extraction
From Text: “Extract
product name and
size from the
following text”
From Images:
“Output parts of the
car with damage in
this picture”
Data
enrichment
“Which country is this
city in”
Sentiment
analysis
“Give me the
sentiment of the
online reviews in
column x”
Content
generation
“Create a customized
marketing email
based on the
customer information
in BQ table “
Video analysis
“Summarize the
content of the video”
● Bring your imagination to your prompts!
● Enforce schema for structured data output
● Batch inference LLM calls are 50% cheaper (vs online inference)
Gemini Model Pricing
Google Cloud Next Proprietary
Google Cloud Next Proprietary
Hands On Lab (Task 1-6)
● Sentiment Analysis
● Extract themes from reviews
● Generate Response to Customer Reviews
Google Cloud Next Proprietary
Demo 1
Google Cloud Next Proprietary
Hands On Lab (Task 7)
● Analyze Images
● Generate Response after analyzing
the images
Google Cloud Next Proprietary
Demo 2
Brand book | Layouts and templates
Layouts and templates
Go to Slido.com
Code:
gfs-aibootcamp
For questions
Layouts and templates
Brand book | Layouts and templates
Thank you!
Any questions?

Scalable ML with BigQuery ML & Vertex AI

  • 1.
  • 2.
    Democratizing AI: Scalable MLwith BigQuery ML & Vertex AI CEO, AdFlex GDE @ Cloud (since 2016) Tu Pham Phuong
  • 3.
    Google Cloud NextProprietary Your AI needs to be fueled by your business data Your business data needs to be unlocked by AI
  • 4.
    Google Cloud NextProprietary Agenda 01 Barriers in Industry 02 Data to AI Journey 03 Built in AI/ML in BigQuery 04 BQML Demo Sentiment Analysis Analyzing Images 04 Questions
  • 5.
    Google Cloud NextProprietary Data and AI Trends 71% Of organizations plan to use databases integrated with gen AI capabilities Of organizations are satisfied with their legacy data platform’s support for AI, indicating there is a lot of room for improvement 14% Data and AI Trends Report 2024 Data and AI Trends Report 2024
  • 6.
    Google Cloud NextProprietary Businesses struggle: Meeting the needs of the AI era 84% Time and complexity from data to insights is long due to fragmented tools not designed to talk to each other business pain point is the need to train and upskill employees in the use of data and AI #1 Data and AI Trends Report 2024 MIT Technology Review Insights
  • 7.
    Google Cloud NextProprietary The Challenge: AI/ML's Perceived Barriers Complexity Expertise Cost 'AI/ML is too complex for our team.’ 'We lack specialized data science skills.' 'Infrastructure and resources are too expensive.' Accessibility 'Finding the right tools is a challenge.’
  • 8.
    Google Cloud NextProprietary Unified Platform for Data to AI Simplifying and unifying data analytics & ML around BigQuery BigQuery Unified Platform from Data to AI Unified experience Single product UX New GenAI powered experiences Collaborative workflows Unified data Unified engines Structured / unstructured Iceberg / Delta / Hudi GCS Cross-cloud (AWS, Azure) Unified governance SQL Spark Python Remote functions Business intelligence
  • 9.
    Google Cloud NextProprietary Typical ML workflow poses many challenges Issues: Streaming/batch data Data processing Export data Manage infra? Learn a new language? Data Governance? Deploy ML model Train ML model (e.g. Python) Where do I host? Multiple products & roles can lead to unnecessary complexity & costs BigQuery (data warehouse)
  • 10.
    Google Cloud NextProprietary BigQuery ML brings AI/ML capabilities to your data platform Issues: Streaming/batch data Data processing Export data Manage infra? Data Governance? Deploy ML model Train ML model (e.g. Python) Where do I host? Simplify with BigQuery ML - data and ML workloads in one place BigQuery (data warehouse)
  • 11.
    Google Cloud NextProprietary Built-in AI/ML | BQML All workloads Why BigQuery? VS ➢ Require every ML use case to go through more specialized systems that require advanced skill sets ➢ Provide ML access to more users through a simple SQL interface Machine Learning for all Built-in ML with SQL ➢ Execute, iterate, and automate ML initiatives all within BigQuery using predefined models ➢ Leverage external models developed in Tensorflow directly from SQL ➢ Export developed models for use in Vertex AI
  • 12.
    Google Cloud NextProprietary Full suite of ML platform capabilities Feature engineering Inference Train model BigQuery Tables or Object tables ★ Rich built in ML featurization functions e.g. bucketize, encode, normalize, resize image ★ Feature transformation modules ★ Generate embeddings ★ Feature store integration ★ Training managed by BQ in BQ infra and in VertexAI ★ Choose from 18 built in models covering major use cases ★ HPO support ★ Batch inference via SQL ★ Import model files (tf, xgboost, onnx) for inference ★ Remote inference via Vertex MaaS or user managed endpoints ★ Streaming inference with continuous queries Tune, evaluate, register models ★ Model monitoring ★ Model evaluation for LLMs and predictive ML models ★ Register to Vertex Model Registry Monitoring
  • 13.
    Google Cloud NextProprietary GenAI using BigQuery ML Vertex AI GCP’s AI Platform BigQuery ML SQL interface to Vertex AI Gemini, 3P and OSS models BigQuery Cloud Data Warehouse Model as a Service (via Vertex AI) 1P Pro, Flash 3P Open Models Hugging Face models deployed via Model Garden can be used in BigQuery ML Via User managed Vertex AI endpoints User Managed endpoints
  • 14.
    Google Cloud NextProprietary Generative AI use cases in BigQuery Entity extraction From Text: “Extract product name and size from the following text” From Images: “Output parts of the car with damage in this picture” Data enrichment “Which country is this city in” Sentiment analysis “Give me the sentiment of the online reviews in column x” Content generation “Create a customized marketing email based on the customer information in BQ table “ Video analysis “Summarize the content of the video” ● Bring your imagination to your prompts! ● Enforce schema for structured data output ● Batch inference LLM calls are 50% cheaper (vs online inference) Gemini Model Pricing
  • 15.
    Google Cloud NextProprietary
  • 16.
    Google Cloud NextProprietary Hands On Lab (Task 1-6) ● Sentiment Analysis ● Extract themes from reviews ● Generate Response to Customer Reviews
  • 17.
    Google Cloud NextProprietary Demo 1
  • 18.
    Google Cloud NextProprietary Hands On Lab (Task 7) ● Analyze Images ● Generate Response after analyzing the images
  • 19.
    Google Cloud NextProprietary Demo 2
  • 20.
    Brand book |Layouts and templates Layouts and templates Go to Slido.com Code: gfs-aibootcamp For questions
  • 21.
    Layouts and templates Brandbook | Layouts and templates Thank you! Any questions?