Oracle's AI Revolution: AI Database, App Dev, Lakehouse & Data Platform Explained

The AI revolution is here and it's changing everything. Within the next 5 years, AI stands to create 78 million

net new jobs and add trillions of dollars in value to the global economy. The core of the revolution is data. Both

your data and the data that others entrust to you. But the question becomes, can you trust AI? Is it safe to

trust the answers it gives? Are the applications it generates ready for the enterprise, secure, correct, and

dependable? You can if you bring AI to your data and use a database where AI and data are

architected to work together. Please welcome to the stage Oracle's executive vice president of database

technologies, Juan Laisa. [Applause] [Music]

All right, thanks everyone for joining today. Uh talk about the AI for data revolution and it is a revolution. So um

as we know our AI is changing everything and it's it's the next big thing in data management without a doubt. Um and it's

not a revolution any of us can afford to ignore. So you know the the quote I like the best about AI is this one says AI

won't replace humans but humans with AI will replace humans without AI. And that applies to enterprises also. And so

um the key to thriving in this age of AI is to transform yourself and your enterprise into an AI leader

uh and use AI to deliver breakthrough insights, innovations, productivity before your competitors do.

And Oracle AI for data is all about helping you do that. U and the great news is AI for data is both easy to

learn and easy to use. And you'll see that u when I describe things today. Now before we jump into AI, I want to talk a

little bit about what I'm going to say about the Oracle AI for data strategy. Uh what we're doing, why we're doing it,

what's different about it. Uh most importantly, how it's going to benefit you and your enterprises.

Um, and I'm going to talk both about what we have today and about a few things that we will have very soon. Um,

and it's important to understand AI is a huge focus for us in the database team. We have over a hundred different

projects going on in AI. We're using AI throughout everything. Uh, it's the most rapidly changing field ever and we're

going to keep uh putting out AI innovations on a quarterly basis. So this is this this is going to be

changing in real time. Uh but before I launch into AI, I want to discuss briefly our overall strategy for Oracle

database and it's pretty straightforward which is we design our products for customers to think and act strategically

um and that understand that gluing things together causes a lot of problems. It creates a lot of costs. It

creates a lot of complexity. We also really believe in open standards and we know that you do also. Uh why? Because

uh open standards are key to creating that competitive ecosystem that benefits everyone. It benefits vendors and

customers. So we've been focusing for many years on a converged architecture and what this means it's an open

standardsbased architecture that uh puts together best of breed support for all the different data types

and workloads and makes them all work together seamlessly. And AI for AI we're doing the exact same thing. We're

putting it all together with everything else and making it work seamlessly. So that's the overall strategy. And there's

going to be three parts to my talk tonight. So first of all, and it talks about what we're believing, how we're

architecting these things together. So we believe that architecting AI and data together is going to enable simpler and

better results. And I'm going to show that. That's the first part of my talk. Then I'm going to talk about uh

architecting AI data and appdev together uh to to enable faster and better innovations everywhere.

uh and innovations that you can trust. And we'll get back to this whole trust topic. And finally, I'm going to talk

about architecting AI data and open standards together. So that's a that's a big deal for us. We're going to be

making some announcements there. Um and again, that's all about enabling AI insights for all your data because it's

open uh everywhere. Okay, so the kind of the three big parts. So let's start talking about how we architect AI and

data together for the our four key products. I'm going to start with AI database, AI appdev, AI lakehouse and AI

data platform. Okay, so first up is Oracle AI database. So this is a new thing. So uh today we launched Oracle AI

database 26AI. So you notice two things here. One is it's Oracle AI database now. It's not Oracle database, it's

Oracle AI database. And we have a new release number 26 AI. It's available now. You can download it today. you can

start using it today. The idea of Oracle AI database is that we've architected AI and data together to create a next

generation AI native database. This is something that we've been working intensely on. And the two key things

that we focus on are the two key AI breakthroughs. One is LLM, which we all know about. The other one's AI vectors,

which is a new data type in in databases, which I'm going to talk about. Uh but in addition to that,

there's dozens of other AI improvements. Now, this new release 26AI is a long-term support release um that's

fully compatible and it replaces Oracle 23AI. So, after today, there's no more Oracle 23 AI. Uh everything going

forward is Oracle 26 AI. And it's very easy to adopt because basically you apply the OR uh October patch, this

month's patch to your database and you're on Oracle AI database 26 AI. It's adding AI functionality on top of our

previous Oracle 23i. That's what it's doing. It's not changing any of the architecture. It's just adding AI on

top. And if you're running Oracle Database 19c, it's also gaining AI capabilities because a lot of the tools

that I talked about today, the AI tools run on top of Oracle database 19c. Okay, so let's dive into it. Um, so first I'm

going to start with AI vectors. Some of you know this already, but it's new to a lot of people. That is the key new AI

data type in databases. So Oracle database now has a new data type called AI vector. And what is an AI vector?

Well, an AI vector is a sequence of numbers that represent the semantic content or meaning of a complex object.

So it could be like a document or an image or a video or or a pattern of data. Uh so this is a brand new data

type enabled by AI and Oracle generates these AI vectors uh from objects using the AI model of your choice. So you hand

Oracle a document or an image. We run the AI model. We generate an AI vector for the object.

Okay. Now that's what an AI vector is. But what good what's it good for? Well um Oracle AI database can both store

vectors which is easy but but more importantly it can quickly find similar objects using AI vectors. So up until

very recently, the only thing that databases were able to do with things like uh documents and images is find

exact exact matches. So the exact same pixels, the exact same words. Uh but if you already know the pixels, then

there's no point in searching for something. You already have it. So what you really want for these complex

objects is the ability to do a similarity search, find things that are similar. uh it like that picture you see

there of the couches, you don't want if you handed the picture of the gray couch on the left, you don't want the exact

same picture back. You want to find the couch that's more most similar to it if you're trying to buy something similar.

And what you can see there is some couches are more similar, some less similar. And the more similar ones have

similar numbers in their vectors. So that's the key the key attribute of vectors. The more similar the numbers in

the vector, the more similar the objects. uh and what we can do in Oracle is we can find similar objects by just

looking at the numbers and we have very fast vector AI indexing so that in milliseconds we can find the closest uh

the closest uh similarity to any document image video anything like that so that's the key idea behind uh AI

vectors in the database and that's the new technology that it brings to databases um now if we want to answer

business questions with AI. What we need to do is find relevant business data in the database and then use the LLM to

answer the the uh questions. And to find relevant business data, you need to do two things. Number one is you need to do

traditional database search and you need to do this new AI vector search and you have to combine them together. So that's

the key and I'm going to be talking about that. I'm going to show you how easy it is to do that now with Oracle AI

database because we have AI architected into the database. Okay. So here's an example uh where we've architected the

vectors and the AI to work together. So the example here is an employee asks a question. So the in in this case the

question is does my dental plan cover braces for my 19-year-old? Okay. So that's an a question written in

natural language. And what we're going to do is we're going to take that question and vectorize it, convert it

into a vector, and then we're going to search our documents in the database to find the closest document that matches

that question. Okay? And then we're going to verify using normal database search that that benefits document

applies to that employee. So that's a simple example of how you combine the new age vector search to to actually

find documents that answer a question and then make sure that the that the uh document is relevant to that particular

user. So it's the combination of the two. Now how hard is it to do this? Well, it's very simple. U here's a

little simple SQL statement that does exactly that. Uh what I I'll read you what it says. It says select the

documents from the benefits documents table where that benefits document applies to

this employee. And then that last line, the order by vector distance says find the closest uh matching document. So

that's all it takes to answer that question in SQL in Oracle AI database. Now, uh, so you see that with one line,

one extra line of that SQL, we're able to answer these complex, uh, questions that were never possible before in an

English language question, finding a document and returning it. And because both the documents and the and the

business data are in the same database, it's completely consistent. Uh, everything is is current and we return

that answer in milliseconds because of our our vector indexing technology. So that means it's good for both OLTP and

analytics because something that returns in milliseconds is very good for OLTP. Okay. So we've we can search across

vectors and all types of of data in Oracle text data, spatial, JSON, XML, you name it, we've we've uh made it work

together with these new AI vectors. Okay, so I just talked about finding the relevant document, but what if you what

happened to the answer to the question? Well, you could go read the document, but it's even better if the if we pass

that result, if we take the result of that of uh searching for that document and pass it to the LM along with the

question, then the LLM can generate an exact answer that you see there in purple at the bottom. And this is called

retrieval augmented generation or rag. And in Oracle AI database, we've automated that whole process. So, you

can either invoke it through APIs or you can run the whole thing as a single SQL statement. So again, that's builtin,

makes it super simple because it's architected into the Oracle database. Okay, let's hear what a few customers

have to say about using Oracle AI vector search. Number one reason we wanted to use

Vector on Oracle database is because it keeps it next to all the other data. It's converged and that allows us to add

the same security policies and not have to have a whole separate system. What we find today is that a lot of our

customers data are distributed across multiple different technologies. Oracle autonomous database helps them to

automate their entire deployment and also bring AI to the data much quicker. We have chosen Oracle autonomous

database because of the ability to use all the different functionality to access data. That means vector search,

select AI, geospatial analysis, blockchain table and so on and so forth. What we found, we compared Oracle

Database Vector Search to a lot of the other options out there. Just the sheer scalability seemed larger. We know it

scales higher than limits that I've heard about elsewhere. And it's faster. It's a great solution. It's cutting

edge. It's scalable. It has the security you need. It just plugs into so much. [Music]

>> All right, that's great. Um, so yeah, thanks. Um, so that's Oracle AI vector search. Um,

that's now we have a an even newer technology now uh that goes beyond that rag uh that we just talked about called

AI agents. So vector searches you hand a question and some data to the LM. The LM answers the question. Now what you can

do with AI agents, you can the agents can take multi-step workflows. So they can plan, they can try multiple

approaches to answering the question, they can use tools, uh they can take actions. So that's the new generation of

AI technologies which we've also built into the Oracle database. So one thing we've done there's a lot the AI agents

is a huge deal across the industry and we're integrating with all the different AI agent frameworks that you see listed

there. Everything from OCI to GCP, Azure, AWS, lots of other things like uh Lang Graph. But in addition to that,

we're also architecting AI agents directly into the Oracle database. So we provide frameworks to easily build,

deploy, and manage in database AI agents and and these in database AI agents will be number one more secure because your

data never leaves the database, number two faster, and number three much easier to use. U so three examples that we have

there, select AI agent, private agent factory, and SQL CL MCP server. So, there's kind of three different

approaches to AI agents in the Oracle AI database. Uh, and now I'm going to have Chris Rice show us how LMS can use the

Oracle SQL CL MCP server to easily answer questions about data in an Oracle database.

>> Standardiz context to an LLM. MCP allows an LLM to answer natural language questions about your database in a very

similar way to a SQL expert, but much faster. The LLM first automatically queries the database metadata to find

relevant tables and columns. Then the LLM writes and execute SQL via MCP to provide an answer to the user. It's like

having a SQL expert that is ready to assist you at any time. Oracle SQL is now accessible via MCP to

enable powerful and simple agentic AI. SQL or SQL CL commands can be invoked by any MCP client such as Klein, Claude or

Copilot. SQL CL MCP server even works for Oracle database 19. Note that we saw the client identified as well as the LLM

in use was identified. These two pieces of information are captured and logged into the database

audit trail. This allows for the DBA or administrator to keep track of which LLM are in use and which clients are in use.

Now, I've further refined it and asked it to summarize the employees by departments.

Now, I'm going to ask it what are the most expensive departments? which salaries are the highest

here. It had to do a multi-table join to come up with the department names and salaries.

Now I'm going to ask it if the location has an impact which again it writes the query and the

query is a multi-table join to come up with the correct answer. And finally, I'm going to ask it to

produce the report in a markdown format so that I can save and share that report with anyone that I need to share this

information with. All right. So, that's a quick demo of the Oracle MCP agent.

It's basically an a SQL expert on demand uh that you can you can use and MCP agent you can download it today this

Oracle MCP agent and use it against any database 19C any any version of Oracle database that works you can download it

for free use it today so it's great way to get started with AI okay uh actually I hit the button ahead of time back it

up please there we go so um we talked about uh how uh AI can can answer questions about data in your database,

but it's not limited to data in your database. It can also combine its own knowledge and things that it finds in

public domain like what by doing a web search to answer questions. And here's an example. You ask how my ABC product

is doing against the market. Uh the AI agents will look through your database, find the answer for how ABC is doing in

your product. then it'll look out on the web, find out how other products are doing and give you a very detailed

answer. So, it's amazing just amazing that that's this is even possible. Uh, and it's actually super si simple to do.

Also, it's not just possible, it's super easy. All right, so I talked about uh generating answers using AI and how we

built that into the AI database. But there's a lot more that we've done with AI. Uh and one thing we've done is we've

also architected AI to help with database management pain points. Uh so just like we have human specialists in

different fields, uh we are also providing specialized AI chatbot assistants to help with different kinds

of database management. An AI management assistant, a security assistant, a knowledge assistant, and a diagnostic

assistant. And each of these uses different data and different strategies to get answers to help you manage your

databases. So it's another giant area that we're investing in. In addition to that, uh we're also transforming data

development using AI. In fact, we're transforming every step in the data uh data development workflow. And I'm going

to go quickly through each of these steps to give you an idea of the benefits of this AI. So first step in

data development is you got to create a schema. You got to go create your tables. And now uh you can easily

generate a schema from a natural language description with the SQL developer AI assistant. You just tell it

what you want. It generates the schema. You can also the next step is a new step which is you need to declare the intent

and semantics of your schema of your tables and columns so that LLMs can understand them. Uh because the LLM if

it can't understand your schema if all it has to go on is cryptic column names or cryptic columns or cryptic comments I

should say then it's not going to do a good job of doing anything. So we've added something called annotations to

the database. So you can basically annotate your tables, your columns, your procedures, tell AI what the semantics

of it are, what what does this data do, what does it mean, when do you use it? U and annotating schemas is a really

important thing to do. So those of us that have been around for for decades in database know that uh things like SQL

tuning was the thing that we used to do. The new version of that in the AI world is annotating your data, describing the

data to the LLM. And again, uh, Chris Rice is going to demonstrate a tool that we've developed that you can use to

annotate your database data and schema. >> We're going to annotate the schema in this database. And to get started, all

we have to do is select the AI enrichment in the tree. It will install the necessary PL/SQL procedures to begin

annotation. These procedures have been backported, so they will work across database 19 and 26 AI.

We're going to switch to a database that I've already begun annotating and creating AI enrichment. As we can see,

the AI enrichment has already begun. There is a group called human resources and there are tables awaiting

enrichment. If we open the enrichment dashboard, we'll see that the schema itself has an enrichment added to say

that this is a combination of multiple things from personnel to accounting to sales to logistics. We also get a

dashboard of how annotated this schema is. There will also be feature functions to import export this for portability.

If we look at the group which is the human resources group, we'll see that the PRS tables actually make up the

human resources schema. We can see here that these columns have a simple description to make the LLM

more aware of what the actual values are. The annotations are free form. So things like aliases or business terms

can also be added. These annotations are a powerful and standard way for the developers to

declare the intent of the data in the semantics. So the LLM is more aware of the data structures and intent of the

data. Here we're making a new group for shipping. These are all our LOG underbar which stands for logistics. But by

adding an annotation, the LLM may not confuse log for server logs, but instead it'll know it's for shipping and

logistics. In addition, of course, the Oracle AI database also automatically provides highle data usage and

information to assist the LLMs. All right, great. So that was creating a schema. Now we've annotated the schema.

We told the AI what you know what the schema is all about. Next thing is to load data. So now you can easily load

data uh with AI assisted creation of ETL pipeline. So again you tell it what the pipeline you want is it'll it'll uh

create the syntax you need to create that pipeline. Um next step well you got to do some testing. So now we've also

added AI assistance for generating synthetic test data that closely matches the production data. All right. And

after that you're ready to go. Now we're ready to use the data. So uh now we have a feature called select AI that lets you

access the data using natural language. And in fact it works with a select statement but the select statement

instead of having SQL has a natural language expression. So you can see that example on the right. It says select AI.

That's a select statement you can hand to the database from any tool. How many employees have not submitted their

benefits plan choice? So from that natural question language question, it's going to go to the database. It's going

to figure that out. It'll return it to you. And then you can add ask follow-up questions. You can say show counts by

department. So it remembers where it was and it can add to that question. Uh and again you can run this from any tool

that that accepts SQL. Um this is also really useful for generating uh uh SQL statements as a first draft for

developers and it does much more than this even. All right. So we've we've created the we we're now uh quering the

data, but there's a lot of different models and a a lot of different parameters to these models. So the other

thing that's important to do is to test and optimize uh the embedding models, the LLMs and

the parameters for your data and your use case. And we've provided a tool for that also called Oracle AI optimizer and

toolkit. All right. So that's the data development workflow. Um, and the other important thing to

understand is all this AI technology is engineered and architected into the database. And what that means is all of

Oracle's mission critical capabilities just work seamlessly. So disaster recovery works, transactions work,

security, analytics, parallel SQL, everything works. Why? Because it's all architected together by us. So you can

just use it and you don't have to worry about any of these things. And this is technology that we've been maturing for

decades. So it's very mature technology that you can run in missionritical databases.

Uh okay. So that was a very brief overview of Oracle AI databases. Um Oracle AI database u as I mentioned

there's dozens of additional AI capabilities way more than I have time to talk about today. Uh for example uh

Exodata AI smart scan uh Golden Gate distributed AI and dozens of more capabilities.

Okay, so that's the first section of the talk, AI database. Huge advancements in AI database. It's really changing the

way databases work. Now, let's move on to AI appdev. Let's talk about what's going on there.

Okay, so this is what we want the future of AI appdev to look like. You say what you want an app to do and the LM

generates the app for you. Okay, it's pretty simple. Uh, we've all seen videos of this happening. It can do this. It

can do this. It can generate apps in seconds with thousands of lines of code. Uh, which is orders of magnitude faster

than we can. Now, what's the catch? The catch is these are enterprise apps that we're talking about and enterprises need

to be able to trust those apps. So, it generates thousands of lines of code. How do you know you can trust that? How

do you know it's secure, correct, dependable? uh you can tell the app, hey, make sure this is correct. Make

sure this is dependable. Uh but you know, you can't trust that either. You can try to put guard rails around it,

but that's kind of trying to fix things after the fact. So, what we're focused on at Oracle is designing trust into the

core of the architecture so that it just works and you can really trust it. And we call these technologies and best

practices that maximize both the innovation and the trust by aring architecting the whole thing together AI

data and appdev. We call these technologies generative development for enterprise or uh gendev for short. And

I'm going to give you a walk quick walkthrough of what we're doing in this space. So the first thing to know about

Gendev is that it's focused on highlevel solution centric languages like SQL. SQL is a solution ccentric language. Um why?

Because enterprises can't trust these thousands of lines of code that no human understands and nobody can evolve.

Nobody could the code is never static. Um and if you think about it, it's a lot easier to understand something that's 30

lines of SQL than thousands of lines of code which does the equivalent of those 30 lines of SQL. That's why these

solution centric languages are really important. So that's what we're focused on. uh so in our agenda methodology we

generate solutions using solution centric languages like SQL and this open application specification uh language

which I'm going to talk about later not so much on traditional code all right so that's the first step understandability

and understandability is key to trusting the generated solution but by itself is not enough there's other risks that we

have to deal with things like data level risks and application level risks so we have to resolve those risks before we

can trust the application. And we're going to start by talking about the data level risks.

All right. So there's three big data level risks that you can have even if you're using SQL. One is uh the

generated SQL can break data correctness. Why? Because it might violate data consistency or business

rules for the data. The second is data evolvability. Um if the SQL depends on the underlying format of the data then

you're not going to be able to evolve the underlying format. So we have to take care of that. And the third and

most important is uh data privacy. Um so we don't want to allow a user to view another user's data. That can't happen

under any circumstances. So I'm going to talk about these three things and and the approach that we're taking and this

is really important. Generating this trust with AI is super important. Okay. So, how does Gendev address data

correctness and evolvability? Well, we do it by combining SQL with these new trusted data APIs to access data. And

the trusted data APIs use this new technology that I've talked about here before called JSON relational duality.

And what that does, it provides a JSON object interface on top of SQL that allows each app to read and write

exactly the data it wants. So uh in that little uh screen you see there that's a purchase order and the API would take

all the data in the purchase order everything and either read it or write it from the database uh exactly the data

that that particular application wants. Now the keys here are that these APIs ensure asset consistency for all the

data in that the API accesses. So the entire purchase order has asset consistency and it doesn't use locks

because if you use locks you can abuse locks and lock up the whole database. Another key thing it does is allows you

to validate business rules on the full business objects. For example on a full PO you can define the the rules around

that and it can validate that on every change no matter what. If you change it through SQL, you change one row, one

column, we can validate the entire business object to make sure it's valid. And the third key is that that data API

doesn't expose the underlying schema. So that enables both app independence. You can write different apps without them

interfering with each other and schema evolution. You can evolve the schema under the app. So again, these

technologies are really key to to architect trust into the underlying database so that you don't have to worry

about AI messing it up. All right, so we talked about those. The next one and probably the most important one is data

privacy. This is a huge risk in the age of AI. Uh why? Because you can't expose other uh users private data. You can't

expose people's medical records. You can't expose their financial records. You can't expose their buying records.

uh you actually go to jail for doing that. Uh but the the the hard part is the privacy rules are very complex and

specific to the role of individuals in each company in each industry. You a simple example here is that if you have

an employee record, well there's different people in the company that can see different aspects of this. Like the

payroll department can see part of the employee record, the human resource department can see a different part.

Managers can see parts of it. So there's very complex rules about who can see what, where, when and how. Um, and the

database has never understood these rules. Um, so and this has been the problem. So today these data privacy

rules are implemented in applications and enforced by custom application code. Um, and this creates two big AI risks.

One is when you're using AI to generate new apps, how do you get it to enforce the rules? You can tell it here are the

rules. make sure when you generate the apps, you enforce the rules, but you can't be sure that it's going to do it.

Uh the second and even bigger risk is that to really gain from AI, you want AI to be able to directly access the data.

Um and the trouble with that is of course it bypasses these application privacy rules. Um and that MCP example

that we just saw was a good example of that. The AI was going directly against the database and accessing data. How do

you know it's not going to reveal private data? So these are two giant risks that have to be resolved in order

to make AI useful in a real enterprise. And the only way this is going to happen, the only way we can fully trust

data privacy enforcement is to implement it down at the source in the database itself. You can't trust the application

because it can be bypassed and you can generate applications that that don't follow the rules. So what we've done now

is we've built into the database very sophisticated rules very sophisticated rules engine uh that can specify exactly

what each end user in a company can which data that end user can access and the database enforces these rules. So

when an end user uh uses SQL to access data or an AI generates SQL to access data for that end user that end user is

only going to see the data that he's supposed to see. He won't see any other data. SQL won't return columns and rows

of things that that user is not authorized to seize. So that whole enforcement mechanism is built into the

database where it can't be bypassed. And I think this is this is absolutely vital for using AI uh in the future without

going to jail because you're you're exposing private data. Okay, so that's kind of data level risk. Let's move on

to app level risks. Uh now there's a lot of app level risks. There's kind of three categories. One is uh the LMS can

be tricked into malicious actions or answers. Another one's LMS can just misunderstand what you ask for. You you

ask for something and maybe what you said is ambiguous, maybe it misunderstands it. Another thing, as we

all know, LMS can hallucinate, can make things up. It can actually make mistakes in generating solutions. So all these

things are are very risky when you're dealing with enterprise data. Now there's kind of two kinds of

applications. One is internal applications that are used by experts that understand the risks and can deal

with them. Uh and there's another kind of applications that you have to have full trust for. For example,

applications that are used by the general public that you have no control over and you can't train. Uh and to to

deal with that kind of application, it requires limiting how AI can be used in the app. And there's two things that we

have to do to do that. We have to restrict the LM's inputs, usage, and results. And we have to validate uh the

reasoning behind the results. That's the only way we can really trust these things for um uh AI to generate these

things. And this is a this is actually a very um it's a very sophisticated area and I'm only going to show an example

here today because it get it can get more complicated. But I'm going to show you how we have a combination of two

things. One's called trusted answer search. Another thing called Apex interactive reports that you can use to

deliver trusted answers to natural language questions. So you you can make sure that the answer you're getting is

correct. Okay, starting with trusted answer search. The way this works is a user asks a question. Okay, that's

similar to what we saw before in natural language. And instead of having the user, the LLM try to answer that

question, what we do is we use vector search to match that question to pre-created reports. So you might have

100, 200, 500 pre-created reports. We match the question to the nearest uh the nearest um report, pre-generated report.

We don't use the LM to directly answer the question. Now you can use LMS to generate these reports or to help

generate these reports or to guide the answers. So natural language question use vector search. We've matched it to a

report. Now Shakiba is going to demonstrate how Apex interactive reports provides trusted answers within the real

time report. >> Millions of Apex applications rely on interactive reports to help people

explore data on their own. They surface trusted information in a way anyone can customize. And now with natural language

support, you can simply talk to your reports. Let me show you. Show me all employees that are managers. One

question and the interactive report applies the filters for me. What's happening here is important. The AI

takes my intent and turns it into report settings I can see and change. It's not generating arbitrary SQL behind the

scenes. Everything is applied as transparent filters, so I know I can trust the results. Group these managers

by department and show me their team sizes. Now the report organizes them by department with each team size right

there. I can even ask imprecise or ambiguous questions like show me high performing managers with a tenure of

four years or more. This is interesting because high performing isn't a column the data. The system interprets my

intent a rating of four or higher and applies right filters for me. I can also customize the appearance and shape of my

report. Highlight those in the sales department in green. And just like that, the top sales managers are easy to spot.

Now, let's pivot to see how departments are staffed. Full-time, contract, remote. This gives me a straightforward

comparison of staffing across all departments. Now, let's go and visualize this. Show me a chart of full-time

versus contract employees. Sometimes I may have a question that goes beyond what interact reports can do

with a single filter, pivot, or group by. For example, which departments have the highest average tenure and within

those, how many managers are rated outstanding? That's more complex. It requires multiple layers of aggregation.

In that case, Apex brings up the analysis assistant. It shows a thinking behind each answer so I can understand

exactly what's happening. The AI still isn't generating SQL. Instead, it interprets my intent and runs safe,

deterministic operations in the background. When possible, it refreshes the report with new chips. This extends

the power of interactive reports to answer open-ended questions with clarity and trust. And from here, I can keep

going asking follow-ups or diving deeper. This is the next chapter for interactive reports in Apex. Natural,

conversational, trusted. And the best part, millions of existing reports can be AI enabled just by upgrading.

>> All right, it's very impressive. Um, it's amazing what AI can do now. And again, this is trusted answers. You know

that these are correct. It shows you what it's doing. Um, and it's limited. It we've limited the LLM, so it can't do

crazy stuff. All right. So, that's trusted answers. Uh, we're also applying this Gendev methodology of uh

architecting AI data and app together to generate full applications. And we're doing that in Apex application

generator, which is a new generation of Apex. Now, what's Apex? Apex is Oracle's very popular lowode IDE that allows you

to visually create apps. Um, now Apex creates missionritical apps that scale to any level and it's very popular. We

have over 3 million Apex apps in production today and thousands of new ones are generated every day. It's also

important that Apex is completely free both to develop and run. So there's no charge for it. Okay.

Yeah, there we go. Um we've rearchitected Apex to be an aa an AI native application generator. Uh

and what we've done is we're rearchitecting Apex to use a new solution ccentric language. Remember I

talked about that like SQL is a solution centric language called open application specification language for creating

apps. And the idea here is this allows AI to focus on specifying what an AI an app should do rather than how it should

implement it. Uh and that makes it much more reliable and much more succinct and like SQL this open appspec language is

orders of magnitude more succinct and understandable than traditional appdev languages.

So the way this works is you just describe in natural language the app functionality, the pages, the data that

you want, the features that you want and AIEX uses the LLM to convert that description, the natural language

description into a succinct and understandable open app specification for the app. And then Apex simply

compiles that specification into a runnable app. Uh, and this generated app will use trusted data APIs, the thing I

talked about a little while ago, to ensure that its access to data is correct, evolvable, and secure. So

that's how this works. It's very different from anything else out in the market. Uh, this is a very simple

example of this open appspec language. This just specifies the simple report. You can see it just says here's the name

of the report, here's the type of the report. You give it the SQL statement and it kind of automatically generates a

screen for all the data uh returned by the SQL statement. Okay, that's a trivial example. Okay, so that was a

quick overview of AI appdev. Um and you've seen what you've seen is the key idea of this Genddev methodology is to

enable ultra fast innovation by architecting enterprise level trust into the stack into the data and application

stack. So it really can't be violated by the LLM. This is an approach that's very unique for us because we're really

focusing on enterprise capable apps. Okay, so that's the second section. Now moving on to the next section which is

open AI analytics and our new AI lakehouse. Um now Oracle's always been about open

standards and we're continue to be about open standards. In fact we're bringing open standards and open platforms to the

age of AI. Um Oracle our AI for data supports all the leading models and frameworks. every LLM, every framework

customers can either call them via APIs or can deploy them as private instances for added security.

So we're open to all the models. Uh we're open to running these things everywhere. So in all the public clouds,

you know, our multi cloud strategy, we can run Oracle database in all the leading clouds. We can run Oracle

database on premises. We can run in this model called cloud at customer where the database runs on premise but it's in a

cloud model and because we've architected all this technology into the database you get to choose where where

you deploy it. Now this advanced AI capability I've been talking about it's not limited to

data in the Oracle database. You can use Oracle AI in any of your data stores uh using Oracle AI databases advanced

federated query capability to create a AI proxy database. So an AI proxy database sits in front of your other

databases and it performs AI action. So a simple example here is you say select AI what were the top 10 products by

sales uh per region that goes to the AI proxy database. AI proxy database understands these other data sources on

prem sources, cloud sources, gathers the data it needs, gives it to the LM. The LM produces answers. So this is great

for uh all kinds of environments where the data is not all stored in one database. And one thing that it's really

good for is running against older versions. So you can put this proxy database for example against older

databases, older uh Oracle databases or older anybody's databases. So that's the AI proxy database.

Uh in addition for many years we've supported uh sophisticated analytics on data in object stores. So we we've

already released things like native delta sharing parallel SQL materialized views all sorts of technologies against

data in object stores. Uh now and this is something we're announcing at this at this uh AI world. We're embracing Apache

iceberg to provide a truly open lakehouse. Um, and for those of you that know what

don't know what Apache iceberg is, it's a standard that's defines the common uh table format in an open catalog for

tables stored in object store files. So you store your data in an object store file in this format. And the whole goal

of this thing is that once it's stored in this common standardized format, then any product can access the data both

read and write the data. So iceberg tables can be both read and written by a multitude of databases and by things

like spark engines and directly by python programs. So it's a completely open standard where all vendors are

interoperable. Um now these iceberg tables have a lot of benefits. Uh they do things like data

versioning, partitioning, basic transactions, basic schema evolution. Uh but these kind of shared tables have a

bunch of drawbacks also. You can't have indexes. It's batch oriented. Uh transactions can't spend tables.

Rudimentary security, slow storage. So what we're bringing to market is a new version of our database, a new product

called Oracle Autonomous AI Lakehouse. And the idea of this is to provide the best of both worlds and resolve a lot of

these trade-offs. Uh so what we're we're delivering is a the vendor independence of a Apache iceberg. So we can read and

write this vendor independent data with all the power of the Oracle AI database. And in fact as of today the 23 AI uh

data warehouses that you already have are automatically converted into AI lakees. So now they're enabled to access

all this data in a lakehouse in Apache format. So with this autonomous lakehouse, we're providing all of

Oracle's best of breed analytics and SQL on top of the whole uh all the all your data lakeink data. So what do you get?

You get the AI vector search I talked about earlier for fast index semantic search uh on all the data lake. You get

the richest analytic SQL of Oracle. So that handles both advanced relational but also graph JSON on the same iceberg

data. uh you get exodata performance because we will automatically cache frequently accessed data in the data

lakeink in exodata and then access it with exadata performance while keeping it completely consistent. Uh you get

scale out data access with a a new technology that's a pay-per-view serverless uh data link accelerator and

you get the secure data access that I've talked about lately that ensures privacy provides comprehensive security

governance and sovereignty. Now along with that you also get a federated catalog of cataloges that that

comes with it. Uh and what that does it provides uni unified discovery and access to data across iceberg and dozens

of other data stores. So it's a catalog of catalog. It finds all your data and catalogs it. Uh and what it enables is

plugandplay access to iceberg data from Oracle SQL. So you just name the iceberg tables and their catalog in the from

clause of a SQL statement and it finds it and accesses it. And you can see the example there where you just put the

schema.t catalog. And this works with any iceberg catalog whether it's our catalog or

datab bricks catalog or snowflakes catalog. So no matter who cataloges the data we can find it we u make it really

easy to access and we can both read and write that data. Okay. Uh, not only can you find that,

you can also load data into iceberg. So, we've adapted our industry-leading ETL and industryleading Golden Gate

replication products so that we can easily move data from your existing systems into iceberg format that's

completely open uh with uh from both operational and analy systems in real time.

Okay. So, that's a big deal. This whole iceberg open data lake. uh in this section what I've talked about and I've

shown is the power and simplicity of Oracle's open architecturally converged everything works together approach to AI

uh and you've seen that this enables uh you to deliver trusted AI insights innovations productivity and apps for

all your data data in Oracle data in other uh stores data in the object store data in data lakes data in thirdparty

databases is um and Oracle one one of the unique things is we're we're running the same

AI across everything. So whether it's operational data or data lakeink data you get the same AI capabilities so you

can standardize and simplify your AI estate and also we're continuing what we announced last year which is all these

AI capabilities that I've talked about are included in Oracle database for free at no additional charge. So you have

these already if you're if you have Oracle licenses. Uh no additional charge. You already own everything I've

talked about. Okay. So that was AI lakehouse. Uh now we're going to transition to our last topic which is

the Oracle AI data platform. And to describe that, I'm going to invite Oracle EVP TK Anan to the stage to talk

about this Oracle AI data platform and how it extends the Oracle AI database that I just talked about by providing an

integrated platform to bring all your enterprise data together and use AI to solve real world business problems. So

please welcome TK Anant. Please welcome to the stage Oracle's executive vice president of AI data

platform, TK Anand. Welcome TK. >> Hi. Hi everyone. Hi everyone. I hope you're all enjoying AI world.

I'm TK I'm TKan. I'm excited to be here and talk to you about the AI data platform. You you you just heard you

just heard Juan Louisa talk about the Oracle AI database and all of the amazing AI innovation that's going on

there and the AI data platform the data platform builds on top of the database and provides an end-to-end platform for

data analytics and AI. All right, let's get into it. So, so, so AI is driving AI is driving the next industrial

revolution and we know it's going to disrupt every industry. It's just a matter of time,

months or years, not decades. The organizations that are going to survive and thrive in the AI era are those that

can augment and reinvent every aspect of their business with AI agents. We've all seen what foundation models can do with

the massive corpus of public domain data they've been trained on, but they know very little about your organization and

your business. The key to achieving AI transformation is to get these models to understand your enterprise data, your

business applications, and your workflows. So the Oracle AI data platform, the

Oracle AI data platform is a comprehensive platform that brings your enterprise data together with

industryleading foundation models to help you build agentic applications and experiences for your business users.

Firstly, firstly, it provides a firstly it provides a data foundation that brings all your enterprise data together

under one roof. This includes all your databases and applications, your structured and unstructured data, your

historical and real-time data. All of this data, all of this data is made AI ready through a unified catalog, through

semantic enrichment, vectorization, etc. And then and then we have an AI platform and then we have an AI platform on top

of that data that en that enables developers to build AI applications and agents. The platform includes foundation

models, AI frameworks and developer tools. We also have agentic experiences for business users to consume these AI

solutions. So, so you can, so you can think of the AI data platform. You can think of the AI data platform as these

two systems that work in concert with each other. Getting your data ready for AI and then leveraging AI to transform

your business. Next slide. Yeah. All right. All right. What is All right. What is the data

platform? Let's look into the key capabilities. At the core of the data platform is an open lakehouse. It's an

open lakehouse built on open standards and there's a unified catalog that brings all your data and AI assets

together to enable integrated security and governance. On top of this data on top of the data we offer we offer on top

of the data we offer the best of Oracle and open-source data engines. You already heard you already heard one talk

about the Oracle II database and that's core to this platform. In addition, in addition, we also offer

Apache, Spark and Flink, which are popular open source engines for working with these data links. And then we have

industryleading AI models like OpenAI. Uh we have AI we have AI frameworks like Langchain. All of these are built into

the platform. And then and then on top of this data and AI foundation, we have an integrated developer workbench that

supports a variety of data analytics and AI use cases. And then we have an agentic user experience for business

users to work with AI agents using interfaces like chat, data visualizations, business workflows, etc.

So, so, so the AI, so the AI data platform, the AI data platform is a brand new pass service in OCI, but it

integrates, but it integrates multiple underlying services together into a cohesive experience. It it relies on it

relies on OCI. It relies on OCI infrastructure services such as CPU and GPU compute for for data processing,

model training and inferencing. It uses object storage to manage all of your structured and unstructured data in the

lakehouse. It it leverages autonomous database. It leverages autonomous database for all the amazing

capabilities that Juan talked about, but most notably high performance query processing and data retrieval for AI

applications, vector store for rag agents that Juan talked about, etc. It it leverages the OCI generative AI

service for all of the foundation models like OpenAI. It leverages Oracle analytics. It leverages Oracle analytics

for its semantic modeling and data visualization capabilities. And fin and finally, the air data platform comes

with a set of built-in services, a set of built-in services for things like the open source spark and flink engines, the

unified catalog, the developer workbench, um the business user experience, etc. So one of one of our

goals one of our goals in creating the A data platform was to ensure that we eliminate that we eliminate the

developer effort required to wire all these products and services together and instead developers can just focus on

building the solution. All right. All right. Now all right now let's talk about let's talk about how

let's talk about how you get your data ready for AI using the AI data platform. The air data platform comes with the air

data platform comes with an enterprisegrade data lakehouse that brings all your data together under one

roof. You can you can bring data from all your databases and applications into the lakehouse. Like I said, it's

structured and unstructured data, historical and real-time data. All of this is made accessible through a

unified catalog. You you can ingest you can ingest data into the you can ingest data into the lakehouse using a variety

of different techniques. You can use batch ETL or ELT pipelines. You can use streaming data pipelines using Golden

Gate or CFKA. You can also just you can also just leave your data in the source system and access it live through the

catalog um through zero copy data integration. And then you can implement you can implement a medallion

architecture on top of the lakehouse which is a pretty common pattern. So for examp so for example all your source

data from the source systems ends up populating the bronze layer. The silver layer is the silver layer is populated

by transforming, cleansing and enriching your data. And the gold layer typically represents the most curated version of

the data that you use for your analytics, your AI applications and so forth. The AI data platform, the data

platform supports open formats, namely iceberg and data lake for managing your data in object storage for the uh bronze

and silver layers. And the gold layer, the gold layer is typically managed natively in the autonomous database for

high performance retrieval within analytics and AI applications. The AI data platform makes it really simple to

it makes it really simple to manage this medallion architecture thanks to the unified catalog that provides integrated

security governance and lineage across all layers. Now, now on top of this data, on top of

this data layer, on top of this lakehouse foundation, the air data platform offers a developer workbench.

It offers a developer workbench where developers can implement all their solutions in one environment. This

includes uh this includes projects like data integration, data engineering, data science, agent development, etc. The the

workbench the workbench consists of the workbench has an AI assisted notebook interface that supports multiple

languages including SQL, Python, Scala and Java. We also have a visual drag and drop interface for low code developers.

The workbench is integrated with git to enable source control, versioning and team development. And you can and you

can use the workbench you can use the you can use you can use the workbench to create you can use the workbench to

create jobs that run on all of the supported data engines. So for example, you might for example, you might run a

you might create a job that runs on the autonomous database or on a Spark cluster or you can create jobs that run

across multiple of these data engines. So So now let's take a look at the AI data platform's developer workbench.

Here's a quick demo. Oracle AI data platform unifies AI models, enterprise data, and an

intelligent developer experience to turn data into value. Meet Jordan, a data engineer at a leading telecom company

tasked with predicting customer churn using customer profiles and customer reviews. In her AI data platform

workspace, Jordan opens a notebook with direct access to her organization's data catalog, which manages all assets using

Medallion architecture. From Bronze Layer's object storage volume, she drags and drops the customer

reviews file to begin exploring the data. With AI Code Assist, she joins customer reviews with detailed customer

profiles, bringing together structured and unstructured data into a single data set.

Next, with just a drag and drop, Jordan instantly receives sample code that demonstrates how to call an OCI Genai

model. This ready to use code gives her a clear starting point. With a few quick edits, she customizes it to analyze the

sentiment of each customer review. Then, she writes additional codes to save the results into a new table in the silver

layer. AI Code Assist can also create code comments and generate full documentation

explaining the code's purpose, logic, and use cases for developers. With profiles and sentiments processed

in the silver layer, Jordan applies a machine learning model from the catalog to predict churn risk. Again, she uses

AI code assist to run the model and save the output to the gold layer, powered by Oracle's autonomous AI lakehouse.

To validate the results, Jordan simply switches the notebook cells language to select AI, pointing directly at the gold

layer. Then she drag and drops the table and asks select AI how many customers are likely to churn. Autonomous AI

lakehouse generates the query, runs it, and delivers the answer. In just a few steps, Jordan has

transformed raw unstructured data into predictive insights by combining Genai, custom ML models, and enterprise data.

Finally, Jordan schedules the notebook to run automatically when new reviews arrive with notifications on completion,

providing end-to-end orchestration with ease. Oracle AI data platform empowers teams to turn data into insights faster,

smarter, and at enterprise scale, driving innovation across the business. All right. All right. Now, all right.

Now, let's talk All right. Now let's talk about Now let's talk about how you leverage how do you leverage all of that

enterprise data? How do you leverage all that enterprise data in the lakehouse to build AI applications for your business?

The AI data platform the data platform comes with a comprehensive set of AI models and frameworks that are ready to

use out of the box. We we have we have we have industryleading we have industryleading foundation models such

as OpenAI, Grock, Llama, Coher and soon Gemini. We have we have we have popular we have popular AI frameworks. We have

popular AI frameworks like PyTorch, TensorFlow, Langchain and Langraph. They're also integrated into the

platform. And these models and frameworks run on highly optimized CPU and GPU compute shapes in OCI. On on top

of these on top of these AI models and frameworks, the AI data platform has an agent studio. It has an agent studio

that's integrated into the developer workbench. The agent studio supports a variety of use cases such as gaining

insights from data semantic search over unstructured documents, orchestrating workflows, monitoring business

processes, raising alerts, etc. So let's see a quick demo of the agent studio in the AI data platform. Oracle AI data

platform reimagines how enterprises build and deploy AI applications. It brings together advanced AI models, a

modern developer experience, and enterprisegrade data security and governance.

Meet Alex, a developer at a soft drinks company. His task, build an AI agent to monitor new government regulations such

as restrictions on packaging or labeling. Inside the AI data platform workbench, Alex creates an agent flow.

He begins by setting the trigger. In this case, a daily schedule that executes the agent automatically.

Next, Alex adds an agent and starts assembling the tools it needs. First, he connects a rag tool to the agent and

assigns a regulation knowledge base that was built with Oracle 26 AI. This allows the agent to scan newly published

regulations, compare them to existing rules, and instantly spot any new proposals. With Oracle 26 AI vectorizing

the content, the agent has precise, reliable access to regulatory knowledge. Then, Alex adds an NL2 SQL tool. This

allows the agent to query the CX and supply chain systems to determine which products could be impacted by the new

rules, whether those involve packaging, labeling, ingredients, or other characteristics. to capture the details

for auditing and analysis. Alex adds a SQL tool that writes a new entry into the company's regulations database.

Finally, he connects Slack and email outputs so the relevant product and supply chain managers are notified as

soon as a regulation is detected. With just a few clicks and minimal coding, the agent is ready for testing.

From the editor playground, Alex can simulate triggers and inputs, review outputs, and monitor the agents

behavior, accuracy, and performance step by step. Once validated, the agent is deployed to a production environment

where it runs automatically on schedule. Just a few days later, the agent proves its value.

When California proposes new limits on plastic bottles, the agent identifies the regulation, logs the details in the

database, and sends a Slack alert. The business owner immediately reviews which products are affected and begins

planning next steps. Agents enable organizations to act faster, stay compliant, and uncover opportunities.

With Oracle AI data platform, enterprises gain a foundation for building and deploying trustworthy AI

applications that keep them agile, competitive, and resilient. So, so the AI, so the AI data platform,

the AI data platform has a unified catalog. It has a unified catalog for not just not just for all of your data

assets, not just your data assets, but also all of your data science and AI models and your AI agents and tools. All

of the AI agents and tools that you build in the developer workbench, they're automatically registered in the

catalog and they support open interop standards like MCP and A2A. But o but over time but o o over over time we over

time we expect most organizations to have numerous AI agents that run in different platforms and applications. So

for example you can have you can have agents running you can have agents running in your SAS applications like

Fusion or Salesforce. You can have agents running within a productivity suite like Office 365. You might have

custom agents that you build uh in a cloud platform like Azure or AWS. But the AI data platform will let you