Skip to main content

AI/BI Genie: A Look Under the Hood of Everyone's Friendly, Neighborhood GenAI Product

Summary

  • Retrieval-augmented generation: Genie combines LLMs with finely tuned knowledge bases and RAG for accurate responses.
  • Query optimization: AI agents orchestrate workflows, optimize SQL performance, and refine responses based on patterns.
  • LLM integration: Genie translates business questions to SQL while maintaining accuracy and performance at scale.

AI/BI Genie: A Look Under the Hood of Everyone's Friendly, Neighborhood GenAI Product

Watch: AI/BI Genie: A Look Under the Hood of Everyone's Friendly, Neighborhood GenAI Product
Talk by Alnur Ali, Software Engineer, Databricks ; Amir Hormati, Director of Engineering, Databricks
Go beyond the user interface and explore the cutting-edge technology driving AI/BI Genie. this video breaks down the AI/BI Genie architecture, showcasing how LLMs, retrieval-augmented generation (RAG) and finely tuned knowledge bases work together to deliver fast, accurate responses. We’ll also explore how AI agents orchestrate workflows, optimize query performance and continuously refine their understanding. Ideal for those who want to geek out about the tech stack behind Genie, this video offers a rare look at the magic under the hood.
Intelligent analytics for everyone: https://www.databricks.com/product/business-intelligence
BI meets AI: https://www.databricks.com/resources/ebook/business-intelligence-meets-ai
Explore the only unified platform for agent systems: https://www.databricks.com/product/artificial-intelligence
See all the product announcements from Data + AI Summit: https://www.databricks.com/events/dataaisummit-2025-announcements

Chapters

FAQs

What is retrieval-augmented generation (RAG)?

RAG combines language models with knowledge bases to provide accurate answers grounded in data. Genie uses RAG to ensure responses are based on your actual data schema and business logic, not generic LLM knowledge.

How does Genie convert questions to SQL?

Genie's LLM understands your business context through RAG, translating natural language questions into SQL queries. The system handles schema understanding, relationship modeling, and complex business logic automatically.

What optimization happens after SQL is generated?

Genie's AI agents optimize query performance, handle execution, and clean results for presentation. They also learn from user interactions to improve future responses and accuracy.

Why is Genie different from a chatbot?

Genie is purpose-built for self-service analytics with accurate SQL generation, query optimization, and business logic understanding. Unlike generic chatbots, Genie provides actionable analytics results grounded in your data on the Databricks Data and AI platform.

Full transcript

0:00 Introduction to Genie Spaces

All right, let's get started. Thanks for coming. Um, so. Uh, uh, and I I work at Data Bricks on the Genie Spaces team. And I'm super excited to tell you today about Genie Spaces. Uh, if you're not actually familiar with Genie Spaces, that's totally fine. We'll unpack that in relatively short order, but the TLDDR for now is that Genie spaces is data bricks product that lets you talk with your data. That's kind of a broad statement.
So to make it a little bit more concrete, you can have the following scenario in mind. Imagine that you work at say an e-commerce company. Uh you might be interested in the following question. What's going on with couch sales? Are they pointing up? Are they pointing down? What's going on? What Genie Spaces lets you do is show up to the product, take exactly that question, what's going on with couch sales, plug it in, and then using generative AI will generate SQL that tries to answer your question, execute that SQL for you, clean up the answer a little bit, and then show it to you.
That's it. No coding required from you really whatsoever. Um, and what we're trying to do or what I'd like to do in this talk is actually give you a peak under the hood of Genie Spaces. So, peel back the curtain a little bit and tell you a little bit more about how it works to a satisfy any curiosity you might have if you're already a Genie spaces user and or b give you information that can help you better design your own Genie spaces.
Cool. Okay. So just by way of an outline uh the talk really breaks down into three parts. In the first part we'll do a quick demo just to level set that'll explain what Genie spaces is. Then we'll give that kind of promised look under the hood and we'll talk about technical details a little bit more. Uh and then at the end we'll kind of conclude with a look back. We'll recap what we've talked about as well as a look forward.
So we'll talk about what you can expect to come out from the Genie Spaces team over the next few weeks, months and beyond. Cool. Okay. So, just a quick intro here. Uh I'm actually not alone. Uh Amir and I are actually both uh on that uh on the Genie Spaces team. Uh Air is right over there. He's also happy to take questions to thank you. Um so feel free to kind of harass either of us at any point if you have any questions.
Uh and we've we've both been in the augmented analytics space for a while. Um, I've built similar products at Microsoft. Amir has done the same thing at Google. And we we've kind of worked on all sides of the spectrum from core machine learning modeling to machine learning infra all the way through to building endto-end product experiences. And we're still banging around in this space because we love it. We think the problems are super interesting.
We absolutely love talking to people like you. And we really do think these technologies are transformative. We think they have the ability to change really how all of us do work. So an important correlary is please get what you came for. Um this session is for you guys. If you want to do like a fivem minute deep dive on vector embeddings like let's do it like please raise your hands, ask questions. Let's make it as interactive as possible even if it's not like on the outline.
So all right so let's level set. So what is a genie space? Okay. So as a running example um let's work with the following simple sales pipeline data model. So there's three tables here. An opportunity table um which really captures the different opportunities say that are in our pipeline as well as different attributes about those opportunities. So say this what stage an opportunity is in. Is it closed? Uh is it not?

3:31 Genie spaces live demo

Um and if it is closed, what date is it closed? And so on. Uh the opportunity table is connected to the account table via the account ID field. Uh the account ID table uh tracks all the different customers that are in our pipeline. Uh the opportunity table's also connected to the user table uh via the owner ID field and the user table contains information about the different salespeople that are working in our pipeline.
Okay, so so far so good. Let's throw it into Genie. Um I guess people can see quick time. All right. Awesome. Um okay. So let's throw that data into Genie. Um the first thing I'll have to do is actually create a Genie space. And that's what I'm I'm doing here. So I open up my data bricks workspace. There's a Genie button. I clicked that. And now I'm brought into this UI. And I I click create in the upper right hand corner.
And now I'm here to build a Genie space. The first thing I have to do is pick my tables. So I'm going to go ahead and pick the opportunity account and user tables that we just saw on the previous slide. Click create. And that's it. Actually, at this point, I've created a Genie space that I can use. I can start asking questions. You can see the tables on the right hand side just as a check. Okay, so it's a sales space.
So, a pretty reasonable question is tell me about the total sales that we closed in one by year. Um, let's let Genie do its thing. And boom, that's the answer, right? Sales numbers broken down by year and a little uh line chart to accompany it. I guess sales went up and down, so maybe that's not too great, but it is what it is. Uh, and then I can also click on show code and see the code that's actually backing uh those results right there.
But What you might have noticed is that it's really not clear these sales numbers how they're denominated, right? They might be US dollars, might be Canadian dollars, might be something else. So, what I might want to do is actually steer Genie to tell me how these numbers are denominated. So, what I do is I open up the task pan on the right hand side here. I click on instructions and I just type in in natural language.
Add USD to any currency column basically. Click save. Copy the same question again. Paste it in. Okay. And that's it. And so now Genie kind of respected what I wanted and I saw the USD label appended to the to the dataf frame header there. Um, okay, cool. So what did we just see? Uh, I actually personally like how Sega put it, which is even if you don't write code day in day out, even if your day job is not to write SQL all day long, you can still use Genie to get what you want out of it.
You can use it to access the data that you want using essentially nothing more than natural language. So that's basically what we just saw. Okay, cool. So let's get into some of the technical details. So what really just happened there? How did we go about answering that question? So as it turns out, Genie is actually a an agentic text to SQL system. That means that there are multiple LLMs and retrieval components all working together to try to answer your question.

7:33 Technical deep dive

Okay, so the starting point is you have to ask a question. Uh so let's say you sit down and type your question. Total sales that we closed in one by year. That's the question we just used. What we do is we grab that question. We send it to another LLM whose job is to assemble all the pieces of existing context that are associated with your genie space and filter that set down to something that's minimally useful. So we don't overwhelm the LLM with context and we kind of focus it on how to best answer your question.
What do I mean by context? We actually will spend a decent amount of time in this talk going through that. But for now the usual suspects are as follows. your data. Your data is a context. What a better way to tell an LLM how to answer questions about your data than to use your data itself. Uh instructions, that's another kind of context. That was what we just saw uh during the demo where I typed in please use USD. Example SQL queries, that's another kind of context.
So these are exactly what they sound like. They're demonstrations. These are pieces of SQL that you write and that you kind of flag to the LLM and say, "Do more stuff like this. write SQL that looks like this. So the job of this component is to actually pull all that context together, filter it down to, as I said, a minimally useful set, uh, and then send it along to the LM. So really the outcome of this component is a carefully constructed prompt.
That prompt, as I said, gets sent over to the main LLM that's responsible for generating SQL. So that LM generates SQL. We're not done at that point, though. Um, what we then do is invoke yet another LLM, kind of an LLM judge, whose job is to sanitize and scrub that answer and take a look at it and ask, do I think it actually I'm an independent third party, you know, unbiased third party judge. Do I think that SQL actually answers the user's question, yes or no?
If the answer is no, then we form a loop. we actually just go right back to square one and we go ahead and determine the relevant context again, hand it off to the LM again, take two, and we kind of keep trying until hopefully we we settle down in something of an equilibrium. So hopefully we've converged at some point on an answer that we think is reasonable. And we're actually still not done because even though we have a reasonable answer, it's not clear how to best present it to the user.
It could be that this particular question calls for text as the right answer modality. It could be that it it requires a table like it could be tabular data. A chart might be more appropriate. So there's another component that basically goes through the results and figures out the right way to present them. And then finally the sum total of all this is that you see the answer to your question. Actually let me pause. Any any Yeah, go for it.
Um, which gives the Could Could you elaborate a little bit? Yeah. Uh, we we don't give you that level of optics mainly because um it's something we like our design. we think it's good, but it's something we're always tinkering with, right? So, I think it would probably not be wise to give you information that then you build on and then just to pull it out from under you. Um, so you don't get that level of detail, but um, yeah, it's it's I also can't necessarily answer the question cleanly because um, as I said, it's it's an agentic system.
So, these agents are talking to each other, figuring out the best thing, the right thing to do. And so, the answer varies depending on the question. It varies depending on the setup of the room. So it's really context dependent but yeah hopefully that helps a little bit. Yeah. So the question was is is the system actually learning as it goes on? The answer is yes in a few ways. One way we'll get to later on. The more obvious way that you might expect is it learns from feedback.
So with it's just like in many other applications. We give you an answer. If you like it, there's a button you can click, thumbs up. If you don't like it, there's a thumbs down. That kind of stuff goes into the system and we figure it out from there. Yeah. So the idea is to grow with you, I think. Yeah. Yeah. What's up? Sorry. Could um so sorry. So in the second box you have mentioned the data. So, are you sending the data to text to SQL LM?
Yeah, so we'll get into that in a little bit. Um, data is sent to an LLM, but you always have the ability to turn that function off, and we always we do our best to respect any kind of security that's been set up as well. So, and is that LLM going out of my network? Sorry, could you say? So, is the text to SQL LLM is out of my network? Is going If you're worried about that, that's something you can disable. Okay. And also do you send the metadata of the tables like the schemas comments and those things also you pass to this text to SQL LLM?
We'll get to that in a little bit but yeah the answer is yes. Thank you. Um so in the first step when it actually when you try to break down all the uh information to you know get the context you probably understand the agent is probably trying to understand what to do. um and it and it'll basically give you a SQL right let let me just assume that I understood all of it even though I did not right for now but on the reflect on answer part your LLM your your agent is basically talking to you know the second part of your you know workflow and also determining if it gave you the best answer right what what's actually happening underneath that and what kind of benchmark does it actually use to say that hey I gave you the right answer yeah so I think the way you can think so the question I think roughly was uh the box in the bottom right corner what what's actually happening when it's talking to these other LLMs and what kind of benchmarks do you use I think the way to think of it is each I mean this is basically I can't say it's an asyclic graph it's a directed graph right so therefore each component gets the information from all the previous components I think that's the way to look at it right so everything you see preceding any ancestor I get all the information from that from those ancestors right Um, as far as benchmarks goes, yeah, I mean there's, uh, it's a great question, there's a number of different benchmarks that we look at.
Some are actually standard and kind of public. Some are proprietary and internal, but we really try our best to make sure that given our user base, we are building a system that really satisfies their needs and not necessarily, you know, any any kind of like public leaderboard or anything like that. That's not our goal. Our goal is to build something useful. I do apologize for this one. I'm going to be annoyingly detailed, but you did say go deep.
Even on something as basic as this, it flashed up incorrect data. You asked by year and it's putting out 2020 20. 5 and so on. So somehow that passed through the judge and even something that basic and that's the thing that we can see. So how can we gain more trust in that? What's going on? How did that fail and how did it fail the judge? So can you talk about what went on there? Yeah, that's a great question. So I think roughly the question was natural language is inherently imprecise.
What happens if I I'm going to ask a question. It's just going to happen whether by design or not of genie that has some ambiguity in it. So what's what's the deal? How do you handle this back and forth with the user? How do you iron out these ambiguities? Uh we'll get to that also a little bit later on. But I think generally the short version is uh we view that as a kind of um as a back and forth with the user. So generally I think our philosophy is when it's kind of clear what a a reasonable answer to your question is we will make that choice but we will also tell you that there are various other ways to interpret your question and we could have answered it these other ways.
If you want us to kind of revisit, go back to first principles and answer your question a different way, just tell us and we can do that. But at least we're not showing you nothing. We'd like to give a best effort attempt at answering your question. And so we kind of think of it as we give you an answer when we can and we work together to kind of refine it from there. I'm sure you'll probably get to it, but the model selection for each of these agents, is that something that's controlled by data bricks or as or how does it work?
Uh, yeah. So, right, it's it's us right now and that's something that we're kind of constantly looking at. My question kind of piggybacks off of the model selection and you mentioned things about not benchmarking anything but I've done tests using lang chains text to SQL tooling and data bicks genie and data bicks genie is obviously gang busters better on the visualization side because that's right out of the box but it tends to be the case that the lane chain integration whenever I connect it with say GPT4.
1 or one of the higherend cloud models tends to outperform it. So to that end, um are there any plans for features where you can kind of select the model you want to use or at the very least have it biased toward maybe a heavier duty model versus a lighter one in the pipeline? Great question. Something we're always looking at. Uh can't really commit to anything at this point, but yeah, it's it's a great question. Uh do you want to continue right now or let's let's actually keep moving.
I feel like these are actually great questions and some of which we're going to touch a few slides down the line. Uh so let's let's keep plugging along and then we can take another pause for more questions in a little bit. Um okay, cool. Okay. Uh so if you noticed carefully in the preceding slide, the determine relevant context bubble was emphasized in bold. So on this slide, let's answer why that's the case. So when we started building Genie, we decided to adopt a better together philosophy, right?

18:23 Context gathering strategies

What does that mean? It means we wanted to put AI and human together to hopefully accomplish more than either side can on their own. I mean, in my opinion, I think natural language is too imprecise. Real world data is just too messy for either side to actually do a 100% good job 100% of the time. The only thing we really ask is that you use the tools we give you to teach Genie how to reason about your data. And the tools we give you range from very unstructured to much more structured.
Okay, so let's take a look at that. If you look at the continuum down here on the left hand side in the on the very unstructured side of the spectrum, we've got instructions. If you think back to the demo I showed in the beginning, that's where we entered the comment about US dollars, right? That was a text instru just free form text. It doesn't really get any less structured than that. All the way on the structured side of the spectrum is what we call metric views.
That's data bricks's own formal semantic model. So again, it doesn't really quite get more structured than that. Backing off a little bit, we've got example SQL queries. Again, these are exactly what you would think they are. their demonstrations to the LLM to say, "Hey, I want you to write more SQL just like this." Backing off even more is your data where here really the hard work is just to annotate your data, assuming you've already kind of curated it in a way that an LM can digest it.
And then finally, backing off even more are suggested SQL queries. So these are not example SQL queries that you write. These are suggested SQL queries that we suggest to you. So, one thing that Genie also does is it sends off an agent to go look around your workspace and mine queries that it thinks are useful for answering your users's questions. And so, we will propose those to you. And we'll see that in just a little bit.
And we'll say, "Hey, do you think this makes sense? Will this be helpful to you to answer some of the questions you've seen?" And you get to give the final thumbs up, thumbs down. Okay? Okay, so we'll talk about some of that uh in a little bit, but given that data is critical, let's start there. And where else to start other than tables and columns. Okay, so what I've shown here is a Unity catalog view of that opportunity table.
If you think back to our data model from the beginning and I've called out and read a few important parts. So the table name, the table comments, the column names, the column data type, the column comments, and any primary key foreign key relationships. So you might ask which of these get sent to the LLM? What does it make sense of? The answer is like all of it. All of it goes. Um, but how? Okay. Um, let's look at the right hand side of the slide here.

21:24 Advanced data intelligence

Uh really what it boils down to is we do a there's a kind of similarity search where for each of the columns in each of the tables associated with your genie space, we smash it together with the question that your user is asking and we just try to figure out which of these columns is actually the most relevant for the question that your user is trying to ask. And there's a machine learning model that kind of makes that determination.
And that pruned list is actually what the LLM gets to see. So some natural uh follow-up questions might be, well, I've got big data. I've got a I've got wide data. I've got a lot of tables, a lot of columns. Is that a problem? Technically, no. Right? That's kind of the point of this uh context gathering and selection component. But really, standard best practices always apply where yeah, if you've got two columns that have familiar sounding names, maybe that's not ideal.
You should probably remove one of those. If you've got a bunch of columns that actually aren't really relevant to answering the questions you would want a user uh to ask in a given genie space, okay, that's probably also not ideal. You should probably remove those columns, too. Okay, so we talked about tables, columns. Let's talk about data values, right? Again, what a better way to teach an LLM how to reason about your data other than by looking at the data itself.
Um, so this really at a high level proceeds in a similar way to the preceding slide where we iterate over a subset of the values for each of the columns, for each of the tables associated with your genie space. We smash those together with a question as well and we figure out okay which which of these values would actually help the LLM figure out how to form good SQL. The real gotcha here is that well big data is big and doing this at scale is not easy and so we've built a feature that we call value dictionaries that basically simplifies this process and does it at scale for us.
Okay, so just to really ram home the point, if you look at this example here, life before value dictionaries kind of look like this, where if you imagine a user is asking a question about Florida, but in your data, you've actually encoded Florida as FL. Uh the LLM might generate SQL that looks like this where it's actually doing a string compare against Florida because that was in the question your user asked. And that would yield an empty result set, right? because you encoded your data as FL.
Whereas if you take advantage of the value dictionary feature, you would likely see the correct answer, which is uh the comparison to the state abbreviation FL. And there's a lots of there's lots of other things we do as part of this feature too. It's not just peeking at the values, it's also making observations about them. So we might say, hey, the minimum value in this column is X. The maximum value is Y. and that could also potentially be useful to the LLM.
So, it's it's all sorts of things. Okay. So, I personally think this is kind of a game-changing feature, but it's new and so it's under active development. Therefore, there are a few limitations that kind of come with it. Uh, one is that we can really only make use of values for currently 30 soon to be 120 columns and that's mainly for scalability reasons. We can really only also make use of let's just call it roughly a thousand values per column.
Uh, and we don't support complex data types. But, you know, these are these are things as I said that we're always looking at and we're always considering revisiting. Okay, so just to rewind, Genie Space is a gentic text to SQL system. Uh the contract that we like to strike is we give you tools. The only thing we really ask is that you make use of the tools we give you for teaching your space how to reason about your data.
There are space level tools like example SQL queries like suggested SQL queries but there also are table level tools. So there's things you might say about a particular table. You could say hey this is useful in this situation. Here's the semantics of this table and so on. Um so on that note let's get into another demo. Okay, so it's our same space and this time I'm going to ask a slightly different question. Top sales in 2021.

26:06 Advanced configuration demo

Okay, because I know the data, I happen to know this answer is actually wrong, this 15 million answer. Um, and the reason is if you think back to our data model, we actually were supposed to do a join. We're supposed to join the opportunity table with the user table. So, how can I tell Genie to fix that? If you kind of noticed over here, I opened up the right-h hand task pane. I clicked on SQL queries, and now I just dump in uh an example SQL query that acts as a demonstration to Genie, right?
And so, if you kind of can see or look carefully, the join is happening right here. Okay. And so artifacts like these really just fall under the same umbrella. I'll just go ahead and give this artifact a name. Top sales in 2021 since that's the question I was trying to answer. They really fall in under the same umbrella as all the previous stuff where I was saying there's a similarity search and we do a kind of matching thing.
The same thing is true here. Um so a practical consequence for you is exact matches are good matches. you know, the closer you can kind of craft these artifacts to match the questions users are trying to ask, the higher your chances of success. Of course, you can't always guarantee that, but just trying to give you a piece of friendly advice. Okay. And so, in this case, we typed in the question, and this is actually the right answer.
So, Genie was able to successfully perform the join in this case. Um, I've got one more demo. Okay. So, um you may have also noticed in the right hand corner here this bubble, right, which basically says we found a few queries that can help you improve the quality of your data space. Um, okay. So, if I if I click the review button, I see this dialogue box here where I'm presented with a bunch of different queries that we've gone off and mined.
So this one attempts to answer the question, what are some closed and one opportunities for each salesperson in October? Here's a fragment of SQL that could be useful in answering that question. What do you think? Good or bad? So I looks good to me. Um here's here's another one. And another one. And there's a fourth one here, too. Um, so I just go through now and I just say, "Okay, the first one kind of makes sense. Um, I'm going to go ahead and click accept.
The others I'm not so sure about. Just turn them down for now. That's fine." And what's happened is we've taken those SQL queries that you've approved and we've added them to your space. Right? So now that query appears here, it's how many closing opportunities are there per salesperson. What we've actually done behind the scenes is dump that into a kind of semantic knowledge store. And so that's also a new feature that we've built uh that sort of houses all kinds of semantic information either minded by us or kind of provided by you that we can use to better answer your questions.
Um and again that knowledge store works at a table level. It also works at a space level. And so if you look at the opportunities table here, uh here's all the columns in that table. You can go ahead and enter additional information about them. You can go to the instructions tab and enter information about this particular table, specify join conditions, um as I'm doing here. You can specify other kinds of semantics too, like synonyms and things of that sort.
So, cool. Um, okay. Uh, let's, you know, we'll just go through the last few slides and then we can actually, uh, we can take maybe questions then. Um, okay. So, what have we seen again? uh Genie spaces is a it's an agentic text to SQL system. Really the philosophy is better together. So try to put human and AI in close contact so that we can do more than either could have done alone. And really the only ask is that you take advantage of the tools that we give you to teach the space how to reason about the semantics of your organization, how to reason about your data.

30:28 Recap and future outlook

These tools range from unstructured all the way to structured to best fit into how you like to work. Uh, probably the least structured tool we have is these instructions again where you can just type in, you know, write things in US dollars. The most structured being metric views, which we did not talk about here, but please find us offline if that's something you're interested in. Again, that's data bricks' formal semantic model.
And fitting somewhere in between are these suggested SQL queries, the ones that I just showed you, example SQL queries, ones that I also showed you, but those are ones that you explicitly author. And your data itself, right? Okay. So, how does this actually matter to you in a very concrete way? Well, again, best practices apply. Uh, keep your space as scoped as possible. Just try to make, you know, have a topic in mind or topics in mind and really try to make sure your space does a good job at answering those questions.
It really is true, additionally, the more you put in, the more you get out. Uh hopefully uh the less structured tools that we give you for providing semantic information are not so burdensome. They lower the bar a little bit. Um and they're not so hard to use, but really if you do use those, it can actually go a long way. So instructions actually help a lot. Example queries can uh go a really long way as well. And you can even consider some of the more structured tools that we give you if you really want to sure a high ensure a high level of reliability.
Okay. So what's on the horizon? Uh really this idea of the knowledge store where we kind of proactively go off and mine snippets of SQL that we think will be useful is an area that we'd like to lean into more. Again, it's really about lowering the bar for you guys and saying, "Hey, this is something that we think can actually help us really do a better job answering your questions. What do you think?" And so, we're that's an area that we're just going to keep leaning into more and more.
General quality improvements. That's always an area of ongoing uh focus, which is really just improving the quality of the SQL that's generated, but also getting into things like a better back and forth with the user to the question uh from earlier, which is, you know, we're natural language is too imprecise. We're not going to get a right 100% of the time, but we can make a best guess effort and then engage in a dialogue with you and say, "Hey, maybe we should have taken we we should have taken a left turn here instead of a right turn or something like that.
What do what do you think?" And then finally, integration. That's a message that we hear from a lot of people, which is they'd love to integrate Genie into line of business applications, dashboards, uh you know, leverage our API. And so th those are all things that we're also looking at pretty carefully. All right, that uh that does it. I'll I'll stop there and we can take any more questions you might have. Thank you for listening.

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.