Your guide to unlocking gen AI’s full

potential with operational databases

Table of contents

Generative AI success starts with your operational data

03
Gen AI makes modernization more urgent than ever
04
Databases are at the heart of enterprise gen AI apps
05
Use case example: Building a product search app
06
Innovate and transform with Google Cloud databases and gen AI
07
Supercharge database development and management with AI
10
Get started with gen AI in your databases
11
Appendix: The Google Cloud database portfolio
13

2
Generative AI success starts
with your operational data
Your enterprise depends on extracting You already know that gen AI is the key to unlocking further
business value. It has the power to transform customer
value from your data. You’re already interactions through improved search and personalized
using operational data—like customer assistance. It can supercharge team productivity by assisting
developers and administrators in their tasks. It can perform
information, financial transactions, and routine tasks to free your staff to innovate and create. And so
inventory levels—to improve processes much more.

and customer experience. And now, In this paper, we’ll show you how to harness the full potential
generative AI has raised the potential gen AI with operational databases—and leverage the next
generation of AI tools to improve employee productivity.
of your data by leaps and bounds.

3
Gen AI makes modernization
more urgent than ever
Leading enterprises are using gen AI in their workflows. And And yet, implementing gen AI isn’t as easy as flicking a switch.
this rapid acceleration of gen AI adoption is creating wider Many organizations are discovering that their legacy databases
gaps in the market—as businesses that embrace it continue to are holding them back from the next level of digital transformation.
move ahead, and those that don’t are left behind.

As an example, integrating gen AI with your operational data Only 14

% of organizations are
enables relevant and real-time responses that today’s
satisfied or very satisfied with
customers value. It’s the difference between a personalized,
accurate response to a customer vs. a generic one. their legacy databases’ support
for AI, indicating there is a lot of
room for improvement.
86% of organizations recognize Source: Google Cloud Customer Intelligence
Data & AI Trends Research, 2024.

that delivering contextual and

relevant user experiences through Lagging technology and poor user experience are just a couple
of the issues caused by legacy databases. Gen AI is bringing new
gen AI integrated databases has urgency to database modernization because the most popular AI
a substantial positive impact. tools for working with vectors, models, and data run in the cloud
Source: Google Cloud Customer Intelligence
and are based on open source database technologies such as
Data & AI Trends Research, 2024. PostgreSQL. With the right tools, you can harness the power of
gen AI within your database to deliver better experiences, drive
productivity, and improve data availability.

Let’s get started.

1. In the financial services industry, Regnology is using

AlloyDB AI to develop a regulatory reporting chatbot.
This chatbot is designed to expedite the process of
Leading enterprises are already using
obtaining accurate answers to regulatory inquiries, from their operational databases with gen
both internal and external users.
AI to improve experiences and drive
2. In the software industry, Linear uses Cloud SQL for its business value in areas such as:
project management platform for cross-functional
teams. They leverage Cloud SQL’s pgvector capability
for similarity-search features that can easily identify Customer support
potential duplicates when a user creates a new issue and
display related bugs that have already been logged.
Marketing automation
3. In the medical device industry, NeuroPace uses
AlloyDB Omni to find electrophysiological features that Product search
are similar across patients with epilepsy in order to help
identify treatment options. They use AlloyDB AI’s Employee assist
embeddings function to transform patient iEEG
(intracranial electroencephalogram) data into vector
representations directly within the database. 4
Databases are at the
heart of enterprise The most powerful enterprise gen AI apps

gen AI apps move beyond generic foundation models

and are built around three guiding principles:
Foundation models are large machine learning (ML)
models that are trained on generalized data. They’re Accuracy
suited for a multitude of purposes, like content generation,
Deliver accurate and up-to-date information
summarization, and simple natural language-based
classification. When developing an enterprise gen AI app,
using a foundation model as a base is more cost-effective and Context
faster than building the gen AI app from scratch. Yet while
Offer relevant user experiences
they minimize the amount of work required to reach
implementation, relying on foundation models also has
limitations. Simplicity
Many businesses find that foundation models alone aren’t Easy for developers to build, operate, and modify
enough for building the contextualized, highly accurate
enterprise gen AI apps required to deliver excellent user
experiences. Enterprises are looking beyond foundation
models—to grounding them in real-time information and
Context
enterprise data. At Google Cloud, we call this “enterprise
truth”—the approach to grounding a foundation model in web Referencing an easily updated knowledge base enables
information; enterprise data like databases and data enterprise gen AI apps to provide responses that are more
warehouses; enterprise applications like ERP, CRM, and HR relevant.
systems; and other sources of relevant information.
Vector embeddings convert text into numerical
The more your business can ground generic foundation representations, allowing a foundation model to understand
models in the enterprise truth that’s specific to your products semantic similarities between words and phrases. Vector
and customers, the more powerful your gen AI app will be. search then enables the model to quickly find the most
relevant information from vast amounts of data.
It’s a really exciting time for
RAG workflows can use these vector embeddings to retrieve
databases, because we’re seeing relevant data into foundation model prompts to refine them.
how organizations can bridge the This minimizes hallucinations, gives more context to the
foundation model’s answers, and provides more reliable
gap between foundation models
information.
and enterprise gen AI apps with
operational databases to contextualize Google Cloud databases include support for vectors, meaning
you don’t need a specialized database. Instead, you can
and personalize the user experience.
streamline your embedding creation and access processes
Andi Gutmans using your regular database.
GM and VP of Engineering, Databases, Google Cloud

Simplicity
Accuracy Any technology relies on the people using it—and so it’s
imperative that your team work as seamlessly as possible with
An operational database stores and processes your data in
your databases and gen AI technologies. Application
real time, making it the most reliable source of up-to-date
developers know and understand operational databases, and
information. And if you aren’t integrating this data in your gen
they can interact with the apps under development. The basic
AI model, then your enterprise app will fall short of its full
framework already exists to make use of your operational data
capability. The retrieval augmented generation (RAG)
in your enterprise application without having to learn an
technique enables you to leverage fresh or domain-specific
entirely new system.
data into your foundation model—opening up new
opportunities to build gen AI apps that deliver answers that are
accurate, informative, and relevant to your end-users. 5
Use case example

Building a product
search app
Customers expect prompt, personalized interactions. And to Now, we have a customer who is looking for some popular
deliver those unparalleled customer experiences, enterprises toys for kids under five years old. Within the app, they can
are relying on operational databases and gen AI to harness interact with a chatbot to answer a wide range of questions
their application data. You can build applications faster with including availability, pricing, and return policies. When the
modern databases compared to using legacy systems. And in foundation model is augmented by RAG, the chatbot can
today’s competitive landscape, this speed is a top priority. answer these questions based on up-to-date information
about inventory levels.
Let’s see how RAG works in a common scenario. In this case,
let’s look at a shopping app for a toy company that uses a So, instead of simply receiving a recommendation for a child
standard foundation model, augmented by real-time inventory under five, the customer would also be given information
and product information from your operational database. about the store closest to them that has the toy in stock.
That’s the type of personalized response that improves the
conversion into a sale.

Pre-step, Here’s a look at how RAG works

Internal data
with an operational database:
Stored to a database 0
through embeddings 0 Internal data is stored in a database through the
embedding model.
App parses natural
language question
Embedding
1 1 Gen AI app uses the embedding model to convert a
model natural language question (“what are some popular toys
for kids under five years old”) to a vector.

Semantic search finds 2

answer from database 2 Embedding model is used to make a semantic search on
the database to retrieve relevant products, and order
them according to stock levels.
Enterprise
Database
gen AI app
3 Database returns the search results to be used as part of
the prompt for the foundation model.

Creates prompt 3
with up-to-date data 4 Foundation model constructs an accurate answer based
on your data, such as “Here is a list of popular toys for
kids under five years old in stock.”

4 LLM

Returns answer
based on your data
6
Innovate and transform
with Google Cloud
databases and gen AI
Google Cloud helps organizations build gen AI solutions and
simplify the management of the databases they depend upon.
With Google’s Data Cloud, data teams can use gen AI tools to
activate their enterprise data and use built-in features to easily
apply AI/ML directly to their data. For instance, built-in vector
embedding capabilities in AlloyDB and BigQuery allow users
to store and generate embeddings within their data stores to All of our databases have vector search
help augment their foundation models and support their gen AI
capabilities. That means you don’t have to
use cases.
deal with complex data pipelines to move
Building enterprise gen AI apps faster your data to specialized vector stores.
Furthermore, you can easily perform filter
Across our database portfolio, Google Cloud delivers world-class
vector embedding and search capabilities. Relational databases
and join operations on your data with your
and non-relational databases alike offer gen AI features to provide familiar database interface. To top it all,
a deeper, more meaningful understanding of your data. you get the required enterprise-grade
data protection, availability SLA, security,
Google Cloud databases are simple to integrate with your
and compliance from your database,
developer ecosystem. They support popular open source database
standards like PostgreSQL and HBase, making it easy to migrate giving you the peace of mind to
from legacy databases. future-proof your application.

It’s also easy to connect your database to external services that

Pranav Nambiar
provide additional AI inferencing services, such as Vertex AI, and Director of Product Management, Databases,

integrate with orchestration frameworks such as LangChain. Google Cloud

Infusing Gen AI across Google Cloud Databases

In-memory Relational Key value Document Analytics

Memorystore Cloud SQL AlloyDB Spanner Bigtable Firestore BigQuery

Vector support 0.6 -0.5 0.9 0.8 -0.3 0.7 0.1

Vertex AI Ecosystem integration LangChain

7
AlloyDB is helping organizations What’s next for AlloyDB
to build gen AI apps
We are continuing to innovate on AlloyDB building the
next generation of AlloyDB AI, which includes new vector
AlloyDB is optimized for enterprise gen AI apps that need
capabilities, easier access to remote models, and secure
real-time and accurate responses. It delivers superior
and flexible natural language support.
performance for transactional, analytical, and vector
workloads. It runs anywhere, including on-premises and on
other clouds, enabling customers to modernize and innovate At Google, we have more than 12 years of experience
wherever they are. innovating on real-world vector algorithms to support some
of our most popular services, including Google Search and
YouTube. We had to invent new ways of indexing and searching
AlloyDB AI is an integrated set of capabilities built into
vectors to meet the most demanding use cases. In addition to
AlloyDB to help developers build performant and scalable
support for open-source pgvector for our PostgreSQL
gen AI applications using their operational data. It helps
databases, we are bringing the next generation of tree-based
developers more easily and efficiently combine the power of
vector capabilities to relational databases. The ScaNN index is
foundation models with their real-time operational data by
a new pgvector-compatible index based on Google’s
providing built-in, end-to-end support for vector
state-of-the-art approximate nearest neighbor algorithms. In
embeddings, and offers:
our performance tests, AlloyDB AI offers up to four times faster
● Easy embeddings generation. With a single line of vector querying than the popular HNSW index in standard
SQL, you can access Google’s embeddings models, PostgreSQL, up to eight times faster index creation, and
including both local models and richer remote typically uses three to four times less memory than the HNSW
models in Vertex AI. index in standard PostgreSQL.

● Enhanced vector support with up to 10x faster

To facilitate easier management of inferencing endpoints,
vector queries than standard PostgreSQL.
AlloyDB model endpoint management makes it even easier to
Quantization techniques support four times more
call remote Vertex AI, third-party, and custom models. In
vector dimensions and a three-times space
addition to Vertex AI, model catalog can also be easily
reduction.
configured for third-party services such as Anthropic and
● Integrations with the AI ecosystem, including Hugging Face.
Vertex AI extensions and LangChain.
Finally, we're bringing two new features in AlloyDB AI to
AlloyDB Omni was built with portability and flexibility in mind. support flexible, accurate, and secure natural language
Customers can take advantage of the technology in AlloyDB experiences. First, we’re enabling gen AI developers to build
to build enterprise-grade, AI-enabled applications applications that accurately query data with natural
everywhere: on premises, at the edge, across clouds, or even language—just like they do with SQL—for maximum flexibility
on developer laptops. and expressiveness. That means generative AI apps can
respond to a much broader and more unpredictable set of
questions. Second, we’re adding a new type of database view
called “parameterized secure view” that makes it easy to
secure your data based on the end-users’ context enabling
you to deliver richer and more flexible natural language
experiences. Together, these advances present a new
paradigm for integrating operational data into generative
AI apps.

8
Vector search across all Accelerating ecosystem
Google Cloud databases support for LangChain

Vector embeddings and searches are critical for building useful Integration with specific LangChain components simplifies the
and accurate gen AI-powered applications—making it easier to process of incorporating Google databases into applications.
find similar search results across unstructured data such as text By leveraging the power of LangChain with our databases,
and images using a nearest neighbor algorithm. Because vector developers can now easily create context-aware gen AI
searches are so important, we provide built-in vector capabilities applications, faster.
across the entire suite of Google Cloud database offerings for
greater operational simplicity and efficiency. You can now store The LangChain integration provides built-in RAG workflows
and search across vector embeddings using your existing across developers’ preferred data sources, using their choice
databases without the hassle of copying data to another vector of enterprise-grade Google Cloud database. Example use
search solution or learning a separate system. cases include personalized product recommendations,
question answering, document search and synthesis, and
● Cloud SQL for MySQL supports exact nearest neighbor
customer service automation.
search and approximate nearest neighbor search.
Developers can store millions of vectors in the same MySQL
instances they are already using and search against their
vector store.
● Cloud SQL for PostgreSQL supports two search
approaches for balancing speed and accuracy. Approximate
nearest neighbor vector search is ideal for large datasets
where close matches suffice, while exact nearest neighbor
vector search is used for precision.
● AlloyDB for PostgreSQL offers high-performance,
pgvector-compatible search that runs vector queries up to
10 times faster compared to standard PostgreSQL. Your
apps can perform fast similarity searches on complex data
types such as text and images, using approximate nearest
neighbor or exact nearest neighbor algorithms.
● Spanner supports exact nearest neighbor vector search on
datasets containing trillions of vectors for highly
partitionable workloads. It can efficiently reduce the search
space to provide accurate, real-time results with low
latency.
● BigQuery supports approximate nearest-neighbor search
on BigQuery data. This functionality is key to empowering
numerous new data and AI use cases such as semantic
search, similarity detection, and RAG.
● Bigtable will soon offer exact nearest neighbor vector
search on datasets containing trillions of vectors for highly
partitionable workloads. It can efficiently reduce the search
space to provide accurate, real-time results with low
latency.
● Memorystore for Redis provides support for vector
storage, enabling ultra low-latency queries for your gen AI
applications. It’s an ultra-low-latency data store suitable for
use cases such as foundation model semantic caching and
recommendation systems.
● Firestore supports exact nearest neighbor vector search.
Developers can perform vector search on transactional 9
Firestore data without the hassle of copying data to another
vector search solution.
Supercharge
database
development and AI can transform how developers operate,

management with AI by solving critical productivity blockers

In a business landscape where agility and responsiveness are 82% of developers spend 30 minutes/day
searching for solutions
key differentiating factors for success, you need to be able to
move fast when market forces change. Database technology is
evolving fast, and database professionals are finding it hard to
stay up-to-date—which hampers both programming quality
25% spend more than an hour searching
for solutions each day

and productivity.

Your operational database is key to managing your

68% encounter a knowledge silo at least
once a week
organization’s data and applications. You want to ensure that
Source: Stack Overflow Developer Survey, 2022
data can flow in and out smoothly, and keep your application
performing well. Managing databases is a job that comes with
a lot of challenges. Many platform engineers, database
administrators, and developers juggle ill-fitting tools, complex
scripts, and error-prone workflows to complete their tasks.

Database Center allows

users to ask ad-hoc
questions on their
database health and
get tailored responses,
ultimately enhancing
productivity.

Gemini for Google Cloud changes that. It provides an AI-powered assistant that helps you be more productive and
creative. It can be your writing and coding assistant, creative designer, migration expert, or even your database
administrator. For database users, it can help you in multiple aspects of the database journey across development,
performance optimization, fleet management, governance, and migrations. Gemini for Google Cloud can help you
move away from legacy databases and migrate your data to Google Cloud databases. It can help your developers,
database administrators, and platform engineers do their jobs more efficiently, and become better at what they do,
with a suite of AI-assisted features for managing, and tuning your database.
10
Migration: Leverages foundation models to assess and Fleet management: Database administrators and platform
convert the schema or database resident code before engineers can manage an entire fleet of diverse databases
migrating data. Easily learn new PostgreSQL dialects, using the intelligent dashboards built with AI, proactively
optimize SQL code, and enhance readability for better assessing availability, data protection, security, and
productivity, easier migrations, and higher efficiency. compliance issues without any custom tools or processes.
With the integrated AI assistant in Database Center,
Development: Developers can build and deploy
database teams can interface with the system using natural
applications faster while meeting security and high
language, making it easier to find the information they need
availability needs with Gemini’s ability to generate,
and troubleshoot problems.
fine-tune, and summarize SQL code with simple natural
language instructions. Data governance: Set data policies to improve security,
regulatory compliance, and control. Manage all your data,
Performance optimization: Operators and developers can
across data silos, in one centralized location. Use built-in
address database performance issues through an
data intelligence tools to check data validity and
easy-to-use interface, providing visibility into all database
compliance.
metrics in a single view, saving time and enhancing
productivity. Database Insights automatically analyzes your
workloads, highlights problems, and provides
recommendations to resolve them.

Get started with Here are the initial steps we recommend

when you’re starting out on your
gen AI in your transformation journey:

databases 1. Look at the possibilities. Gen AI is rapidly reshaping the

landscape—and implementation can be the difference
between falling behind or leaping ahead. Research what
Gen AI is driving a new wave of database
your competitors are doing, and get inspiration from how
modernization—and it is accelerating rapidly. other organizations are using gen AI.
And organizations who fail to modernize will find
2. Put together a development team. Then, align your key
themselves on the sidelines.
decision makers on your goals to move forward. Consider
augmenting your existing team with assistive
Modernizing is a crucial strategy, but moving from legacy technologies to lighten the load. For example, Gemini for
systems can seem like a daunting prospect—as for many, the Google Cloud can assist with database management, so
road to modern systems can seem unclear and intimidating. you don’t need to hire specialist database administrators
and platform engineers.

3. Start small. Make simple scenarios and create use

cases, such as cleaning up your support ticket queue.
Gen AI can identify duplicate support tickets for your
team, or pull out previously solved tickets that are similar
to the one a support person is looking at, to provide
precedent and guidance on what to suggest.

4. Search for opportunities to improve. This might

include automating tasks like maintenance and
background business processes, or personalizing the
customer experience so your users get a better, more
fulfilling interaction with your organization. Think about
your operational data, and how you can use it to add
context and relevance to your application.

11
Start your
transformation
with Google Cloud
We’re here to guide you. With our years of
experience in implementing these systems for
ourselves and others, we have a solid
understanding of the challenges and
opportunities you’re facing.

To help simplify your modernization journey,

Google Cloud offers a database modernization
program that combines the best databases,
migration tools, expert guidance, best practices,
and financial incentives. It’s designed to help you
move from Oracle and SQL Server to Google Cloud
databases, helping your organization meet its
gen AI goals.

Contact us to talk about migrating your

database from a legacy system, developing
a new application, or simply finding the best
way forward for your organization.

Talk with a database specialist

The Google Cloud
database portfolio
Google Cloud provides an intelligent, open, and unified data and AI cloud to support your
gen AI future. Revolutionize customer experiences with operational databases you know
and love, in virtually any environment—whether in the cloud or on-premises.

● Cloud SQL is a fully managed relational database ● Bigtable is a highly performant, fully managed NoSQL
service for MySQL, PostgreSQL, and SQL Server. database service for large analytical and operational
workloads. It offers up to 99.999% availability, and
Database Migration Service helps simplify
processes more than seven billion requests per second
●

migrations from legacy MySQL, PostgreSQL, SQL

at peak, with more than 10 Exabytes of data under
Server, and Oracle databases.
management. Migrate from databases like HBase
● AlloyDB for PostgreSQL is a PostgreSQL-compatible and Cassandra.
database service for your most demanding enterprise
BigQuery is a fully managed, AI-ready data analytics
workloads. We also offer a downloadable
platform that helps you maximize value from your data
edition—AlloyDB Omni—designed to run anywhere: in
and is designed to be multi-engine, multi-format, and
your datacenter, your laptop, and on any cloud. Use
multi-cloud.
AlloyDB AI to easily build enterprise generative AI
applications, and simplify migrations to AlloyDB with ● Firestore is a highly scalable, massively popular
Database Migration Service. document database service for mobile, web, and server
development. It offers rich, fast queries and high
Spanner is a cloud-native database with virtually
availability up to 99.999%. It has a thriving developer
●

unlimited scale, global consistency, and up to 99.999%

community of more than 500,000 monthly active
availability. At peak performance, it processes over
developers.
four billion queries per second. Migrate from
databases like Oracle or DynamoDB. ● Memorystore offers fully managed in-memory
Redis and Memcached service that offers sub
Bare Metal Solution allows you to lift and shift Oracle
millisecond data access, scalability, and high availability.
workloads to Google Cloud.
Memorystore for Redis Cluster is a fully managed service
which can easily scale to terabytes of keyspace and tens
of millions of operations per second.

13