Sponsored by: Tredence | THE AGENTIC ENTERPRISE: ONE YEAR LATER From Vision to Value in CPG

Summary

This video shares Tredence's one-year journey implementing agentic AI for CPG clients — moving from pilots to production — and the central role of a Context Graph in preventing agents from hallucinating on business data.
The Context Graph, built on Unity Catalog's business glossary, resolves KPI ambiguity at enterprise scale by encoding multi-region definitions, fiscal year context, and Power BI semantic layer integration into a domain-specific knowledge graph on the lakehouse.
A multi-agent software development lifecycle framework covering code generation, testing, documentation, and code review has delivered measured sprint velocity improvements, and the video covers feedback loops and custom reward models for continuous agent improvement.

Chapters

FAQs

What is the Context Graph and why does it matter for agentic AI?

The Context Graph is a domain-specific knowledge graph built on Unity Catalog's business glossary that gives agents the enterprise context they need to stop hallucinating on business data. In this video, Tredence demonstrates it resolving a supply chain KPI ambiguity — a lead-time definition that differed by region — which had quietly been costing the business.

What does it take for a CPG company to move from agentic AI pilot to production?

Based on Tredence's experience working with CPG clients, the key bottleneck is not the technology but the context layer — agents need structured business knowledge to answer correctly and consistently. This video argues that investing in a Context Graph on top of your lakehouse is the foundational step required before agentic AI can scale beyond proof of concept.

How did Tredence improve software delivery velocity with a multi-agent SDLC?

Tredence implemented a multi-agent software development lifecycle framework covering code generation, testing, documentation, and code review, which the presenters report delivered 15–40% sprint velocity improvement with fewer defects. This video explains how human-in-the-loop design was essential to achieving those gains responsibly at scale.

How do you handle KPI ambiguity when deploying agentic AI for enterprise clients?

This video shows how different business units, regions, and fiscal calendars can define the same KPI differently, causing agents to produce inconsistent or incorrect answers. The Context Graph addresses this by encoding multiple definitions with their contextual scope, allowing agents to resolve ambiguity before generating a response.

Full transcript

[00:07] Welcome. Good afternoon. Uh thank you for taking the time out. Um What we wanted to talk about like a little bit of introduction first. My name is Majaz Mohammad. I am I lead the CPG business for a company called
[00:19] Tridens. For those who do not know Tridens, it's a data and AI company. We work with eight out of the top 10 CPGs in the world, nine out of the top 10 retailers in the world. Right? We power their data and AI strategy.
[00:32] Um and I have with me Arnab. Uh if you can give a quick introduction about yourself, Arnab. >> My name is Arnab and I head data and AI engineering at Tridens. I've been with the firm for the last 9 and 1/2 odd years, you know, grappling with
[00:45] different kind of technologies and very, very, you know, deeply into the context layer and and the, you know, enigma around it. Happy to explain more as we as we talk. >> Awesome. So, I think initially this session was supposed to be about
[00:59] a particular company. It's not about the company, right? There were some logistical challenges to to get them here and all that, but it's not about that anymore, right? Like what we
[01:11] are going to talk about is our experience in general working with CPGs, right? And especially last year we had given a presentation on agentic AI, whether it's a hype or
[01:24] reality. Right? What we want to talk about is our journey from last year to now. We have done some implementations. What are the challenges we have faced? What can you learn? We make it as practical as
[01:36] possible for you, right? And with with specific examples of what we have seen work and what we have seen as problems in the agentic AI space. Right? And we will keep
[01:48] the last few minutes for questions, right? If you guys have any, I don't know if you have some mics that we can, um, you know, um, rotate around. Um,
[02:00] fantastic. We have We have a mic that we can rotate around for questions and all, right? So, last few minutes we'll we'll keep for questions. But, uh, um, I think I'm supposed to use this. What we want to talk about is the again
[02:14] the promise of agentic AI. Of course, everywhere you're going agentic AI is what you hear. In fact, I saw a meme, a video, right, where it said, "Every
[02:26] uh, graduation speech ever." Right, or every keynote speech ever. And the guy comes on the stage and he says, "Agentic." And leaves. That's all. Right? So, that's what it has become
[02:38] this year. Everyone's talking about agents, right? Now, is it reality, or are we stuck in pilot forget burger pilot, you know, jail, let's say, right? Um,
[02:50] and then the more important learning that we have had over the last year is we have built agents, but those agents, because they don't have the right data definitions, the the
[03:03] right right data, you know, structures below it, they start hallucinating quite a bit. So, the most important thing we have found is the context. So, we'll talk about that context. And I think in the agentic AI's uh, the the curve of
[03:16] adopting technology, that that's where we are at the peak of the hype right now. Right? We might go down and then stabilize a little bit after that. Right? So, with that, uh,
[03:28] let me just summarize what we talked about last year versus what we want to uh, talk about here, right? First of all, the vision of agentic enterprise is what we had painted last year.
[03:40] We talked about the potential of agentic AI and laid out a theoretical framework of how agentic AI might be adopted within industries, especially in the retail and
[03:52] the CPG industry. Now, what we have done is we are actually uh we have worked on creating agents. We have worked on deploying agents into production. So, we want to talk about
[04:05] our journey of turning that vision into reality. And we want to talk about the lessons learned, the value delivered, etc. Right? Of of of this whole agent AI.
[04:18] Uh The biggest lesson, like I told you, right? We have created agents. Uh we'll give you examples of what what those agents are, what we have created. The big biggest thing that we have found
[04:31] is the context gap. Right? Your data foundation even though it is solid, right? The contextual layer, and here also there are multiple things, right? Contextual layer when I'm talking about
[04:43] there's a company whose fiscal year is from July August to July. The user goes in and starts typing a question, "Hey, what were my sales for the last fiscal?" Agent doesn't
[04:55] understand what fiscal means. That's context for you. Right? And then um then there is the connections. Like there is one more component of that, which is the different definitions. What
[05:08] do you How do you define sales? Different companies, even within the company different regions have a different way of defining what sales means. Right? Do you include your discounts and rebates in it? Do you not include your
[05:20] discounts and rebates? Like there's variety of definitions of what sales means, right? And when the agents look at your data model and see these different definitions of one particular KPI, they start getting
[05:32] confused and they start hallucinating. Right? That's where we have seen issues. Now, without the uh context layer, right? So, I mean, that's the uh image that we tried to show you here.
[05:44] The bridge between your AI brain and the enterprise data is the context layer that that uh uh you kind of have to create.
[05:57] Sorry. Let me let me Yeah. Yeah, so wanted to give you a few examples, right? Example number one. Let's think about uh the entire value chain of a manufacturing company, let's think,
[06:09] right? You're doing sourcing from the raw material sourcing raw material from your suppliers, bringing it to the plant, at the plant you're doing the manufacturing, moving it to a warehouse,
[06:21] and from the warehouse you're delivering it to a customer. Now, in this case, there are multiple lead times as such. Right? Getting the product from your supplier to your raw
[06:33] to your plant is a procurement lead time. Manufacturing within the plant itself is your manufacturing lead time. Moving it from manufacturing plant to the warehouse is your uh you know, transit
[06:45] time or the transit lead time in there. Shipping, picking, packing, shipping from the warehouse, delivering it to the customer is another lead time. So, if you just go to the I mean, all these definitions exist in your data
[06:58] model as lead time. Right? And uh if you go if you create an agent on top of this data model, and you go and ask a ques- a simple question like, "What's my lead time to
[07:10] uh you know, to deliver a product to the customer?" Now, the agent is confused which lead time you're talking about, right? If you don't have the proper context layer. That's where I think uh we have run into some problem.
[07:22] Initially, we started creating agentic agents based on our last week last year's vision. And then we started seeing all these hallucinations come up, and when we dug deep, what we realized is th- this is one of the problems. Similarly, I don't
[07:35] know if you can talk about one of the uh sales related problems that we have >> Yeah, yeah, so that's that's pretty simple, actually. If you look at the CPGs, CPGs are mostly organized by markets, right? You got MEA, NA, you got
[07:49] Middle East, and you got Africa, etc. Right now, um say for example, you look at net sales, right? Net sales may be, you know, defined as, you know, uh your your revenue your your total revenue minus, you know, minus the
[08:01] discounts, minus, you know, the returns, the discounts, etc., right? For for the North America geography, in in Europe, they may not consider discounts, right? So, it's it's defined in in both places in different ways. Now, how do we get our get our arms around it and
[08:13] understand, right? When the question is asked by someone in North America, it needs to refer to one definition of of net sales versus the other, right? That clarity is generally provided by the
[08:25] context layer. Right? >> Yeah. So, I think in that example, again, earlier what we had talked talked about, all right, net sales, do you just think about it as North America
[08:37] wants to include rebates, discounts, and consider it as net sales. EMEA, they don't consider rebates, they just consider discounts, and they that's net sales for them. And when a user from
[08:50] uh either region, you know, queries an agent, "What were my net sales last year?" And in this case, last fiscal, let's say, the agent is confused on two different levels now. What is fiscal?
[09:02] What is net sales? Right? And that's where it starts again being the confident agent that it is, it'll start hallucinating. Right? And your accuracy starts going down. So, that's kind of the problem
[09:14] that we have faced. So, first of all, what we wanted to do was define what context means, right? And what is How do you build the context? And we'll take these two examples throughout our journey over here. One is your lead
[09:26] time, the second one is your net sales. Now, context, first of all, is the definition. What does the metric or the entity actually mean? Right? And then there is the
[09:39] relationships. How is it related like in in my case when I talked about the value chain from your suppliers to your plant to warehouse to the customer, all those connections, right? How do you
[09:51] define those connections within your data? That becomes the second part. Then the usage. Who uses this data and for what decisions? Now again going back to my example of the lead time, if there is
[10:03] a procurement person asking, "Hey, what's my lead time?" They are probably talking about the procurement lead time from the supplier to the plant. Right? So and the agent needs to have that context of
[10:15] hey, who is asking the question? Which definition should I choose? None of those definitions are actually wrong, right? And then um the purpose. Why does this data exist in the enterprise, right? What's the
[10:27] purpose of the data? How is it used and and those type of things. Lineage. So once your definition of the uh you know, specific KPI is clear, which tables do you need to source it
[10:39] from? Right? That becomes the uh more important question. So lineage not only in terms of lineage in your gold layer of your database, but also where does it get populated from? What are the source
[10:51] systems it's coming from? So that the agent can understand the the source of your uh data. And then scope. Again, which business unit, geography, you know, time period
[11:03] does it apply to? So um in your net sales example, Anup, if you can walk through each one of these definitions, right? Like each each one of these entities. >> Yeah, I'll >> What what does it mean?
[11:15] >> So so for example, right? I mean I'll I'll not I mean of course the definition is specifically how do you define sales and we talked about two kinds of definition, one for North America, one for one for EMEA, etc. But when you look at scope, right? It
[11:28] again becomes, you know, is you know, which business unit or geography? Is it is it North America querying for the answer? Is it Is it you know Europe querying for the answer? And then when you're looking for lineage, which product where did it sell? Okay, who
[11:42] sold it right from the ERP flows from your SAP systems through your bronze, silver, gold layer of your medallion architecture, how does that entire data flows through, right? And where does it flow through from there as well if there are reports etc. supporting it. So that
[11:55] and then if you're looking at relationships, it's like sales to products to customers to you know to orders. So all these could be your entities which you which you'll have to track and and these are the entities which influence sales at the end of the
[12:08] day. And then of course usage and purpose purpose are actually self-explanatory. >> Okay. Awesome. >>
[12:20] graph is in place, how that solves the problem. Right? I mean, first of all when we are talking about context graph here, there are few terminologies we should be aware of, right? One is called ontology.
[12:35] The second one is knowledge graph and the third one is context graph. So maybe Arnav you can tell us the difference between the three before we move forward. >> Okay. So ontology is nothing but you know it's
[12:48] it's all the nodes, edges are different entities that connect your businesses together, right? Typically it's empty, right? But it shows all the relationships that are there in your in
[13:00] your in your enterprise. Okay, so it could be from the source systems, it could be up right up to the reports etc. but at a entity relationship level, you know what are these edges, what are these nodes. But there is no data or metadata I mean no data metadata
[13:13] associated with it. Now a hydrated one becomes a knowledge graph, right? You hydrate it with data to a certain extent. You don't want to do an entire copy of your data because you don't want you already have the data in your lake,
[13:26] right? Ontology can solve at least 80 to 90% of your questions. 10% are, you know, actually routed to your knowledge graphs. So, that is your knowledge graph. And what happens is, over time, your knowledge
[13:39] graph becomes huge, right? Because there are so many entities, so many traversals, whether you do a depth-first search or a breadth-first search, etc., right? It becomes very cumbersome. So, you this understand for each of the questions you
[13:52] you create small pathways or or or subsections of the knowledge graph, which is known as the context graph. So, that that way, from a latency perspective, if you're if you're if you ask a question, you'll get the answer as
[14:05] as soon as possible. So, that's these three layers you'll have to be mindful of. And then, you know, there is of course an entire methodology around how do you hydrate, how do you create, etc. So, we can go into those as as and when
[14:17] we talk. >> Yeah. Awesome. So, I mean, the the analogy in my mind is I'm a supply chain person at the end of the day, right? So,
[14:31] your ontology is like an empty warehouse with no products inside, right? The the bins are there, the structure is there, but products are not stored in the warehouse. Your knowledge graph
[14:43] is a filled warehouse, right? Now now you have actually filled products on the shelf and and everything. So, that becomes your knowledge graph. And your >> context graph. >> Context graph is one
[14:56] section of that particular warehouse, right? Like maybe you're going only to the soaps section. Right? So, that becomes your context graph, basically. And that's where it's the the context, etc., is stored. So, now,
[15:08] going back to our example, let's say we have put all of these three in place. Obviously, we've created the ontology, hydrated with all the data. Your context graph is ready. That thing is ready with the lead time. All the three four
[15:20] definitions of lead time are in place. How does that solve the problem? The agent first queries the Unity Catalog's business glossary to find the three four definitions of lead time that we talked about. Right? Then the context
[15:34] graph identifies the user's domain. Right? And here it can be okay, the user is a procurement person, the user is a manufacturing person, the user is a logistics person. Based on that, it picks which definition to use.
[15:48] Right? And then the lineage tracing tells it which table to go to to query for that data. So, using these three mechanisms, it will come back with the right answer. Right? Um no hallucination, no
[16:02] ambiguity. The same thing in the sales uh example, how do how does that work? >> Yeah, it's it's it's it's pretty similar, right? I mean, you will you'll of course through Unity Catalog, etc., you can understand who is querying the
[16:15] data, right? Or from which region are the other people querying the data from. Then you look at the context graph and understand what the domain is and for which your KPI is defined, right? So, example in net sales, you know, whether to include discounts, whether to include
[16:28] anything else, right? Based on which market it it is being from it is being from queried from, and it selects that, and then of course traces the lineage, constructs the query, executes the query, gives an answer if it's a
[16:40] deterministic one. If it's a non-deterministic one, then of course it can give a set of answers. We'll have to look at how to rank, prioritize, and and and and show. >> So, if the user is from you know, uh North America, they ask the
[16:54] question about net sales, it figures out the right definition that hey, I need to include rebates, discounts, the entire thing. Give it give the answer in terms of net sales. The user's from EMEA, it picks the right definition in which only the discounts are included and no
[17:07] rebates, right? And and then it comes back with the answer. So, that's kind of uh how it worked. Right? And then the agents the different so for for
[17:19] accelerating the the development life cycle. Right? Off not just the context graph, but the entire thing. We have developed some agents to accelerate the software development
[17:31] life cycle. Right? And we call it AILC nowadays. So maybe you can talk about some of these agents over here that that that we have developed. Right? Starting with the code generation agent.
[17:43] >> Yeah, sure. I mean from a from a code generation standpoint, right? When we when we look at it is, you know, if you're doing a migration or if you're you know, developing your end-to-end end-to-end you know, bronze, silver,
[17:55] gold layer. Of course, you know, what we've done is we've created you know, if you've got PySpark code etc. or you're converting code from an Informatica or a Teradata onto onto Databricks. So what we've done is we've
[18:08] we've created you know, code generation agents which basically has LLM as a judge also built into it which basically takes your Informatica XMLs, your whatever's there, right? Or you know, if you're trying to get for example, for
[18:20] hydrating or or or creating an ontology, you need the metadata from a Power BI table or a or a you know, and you've got a Power BI semantic layer. So you we've got what we've done is, you
[18:32] know, you can put the Power BI semantic layer to the agent as an input. It is going to basically pass through the information, take out the relevant metadata, and push it onto your knowledge graph. So so that's one one point of it. And of
[18:45] course of course after that to test our agents etc. we've created testing agents test our entire knowledge graph. We've created testing agents. So what these testing agents basically do is
[18:57] what we've done is we we generally go there and collect set of competency questions from our users for which they also provide answers. So this test testing agents basically, you know, tests the knowledge
[19:10] graphs against against the different entities and the answers that have been provided by the users, and and ensures that the knowledge graph or the ontology, etc., is valid, so that we can scale scale out or scale out the the entire agents, right? And also for for
[19:23] documentation, we've got also documentation agents, which basically, whether it's a you know, whether it's a Jira or a Confluence, etc., from where you want to want to collect information, so that you can make your context more and more
[19:35] richer, so that, you know, it it becomes much much more meaningful. So, we've got agents to be able to crawl through that and and create a structure which can be pushed into a Unity catalog. So, these are the few agents that we've created.
[19:47] >> Awesome. Thank you, Arnab. And as a result of these agents, right, we are able to deliver, you know, the code, the programs approximately, I would say, on an
[19:59] average, about 30% faster, especially when you look at the entire velocity of, you know, user stories that are being developed. Obviously, if you think about just the coding agent that accelerates your coding velocity quite a bit, like
[20:13] it it will be 50-60% acceleration. But then you you have other things to think about, like your approvals that you have to go through, right, your you know, the testing that you have to go through, and then the iterations that
[20:26] you have to do, and all that, that brings it down to about 30% acceleration on a realistic basis. We have done this, deployed it in in production and and everything at large scale, we have done it, and the the number that
[20:52] down what that 30% means. Sprint velocity is uh uh you know, compared to the baseline, about 15 to 40% improvement is what we have seen in various deployments that we have done.
[21:04] Defect rate has gone down 15 to 25%. Right? And finally, your code review turnaround, earlier it used to be two to three days. It's like less than one day right now. Right? In terms of especially
[21:17] if you're using code review agents, etc. to review the code and and point out issues in there.
[21:29] Madam, you want to take up challenges and lessons learned? >> Yeah, so I mean I can, right? So so let's let's talk about it, right? I'll you know, there were a few few things that that we've mentioned over here, right? For example
[21:41] hallucination, right? I mean important thing, of course, we did talk about getting the context graph in, right? But that reduces your hallucination to an extent, right?
[21:54] You're just placing a context graph or a or a knowledge graph or a context graph in does not is is not good enough, right? You'll have to ensure that it is up to the mark, right? For that, a
[22:06] knowledge graph goes through a lot of cycles, right? When it is generated, right? So for example, uh if you're if you're gathering metadata from if you there are three types of knowledge in an organization, right? There's tribal knowledge, which
[22:20] is in someone's head. There is there is knowledge which is in your Jira and your confluence and your ERP process diagrams. And there is knowledge which resides in your data lake, right?
[22:32] The idea is to you know, grasp all this knowledge, right? And create a single metadata repository. The more richer you make it, the better it is, right? And then, of course, the context graph becomes
[22:45] narrower and narrower as we as we go along. But it's important that what the first ontology that you get, okay, when you you know, get all this information synthesizes all this information and you're able to create a empty, you know,
[22:57] on the ontology, it's it'll never be up to the mark. What you'll have to do is you'll have to go through a process of pruning, okay, straightening it and reducing circular references, pruning, etc., looking at it hard, right? And then
[23:10] you'll be able to see that, you know, you're you're able to reduce it quite a bit. Then you'll have to populate it with, you know, some some constraints, etc., which you which will be there in your lake etc. lake. And once you do a
[23:22] second round of validation, you'll see you're able to prune it further. So, you'll get a much narrower sliver of your ontology, which is which becomes much more traversable. Important thing is this has to be right for
[23:34] hallucination to reduce significantly. The second one is, of course, human in the loop, right? All these needs expert validation, especially you've got, you know, domain experts, right? There are cases when where we have been we have we
[23:47] have we have been talking to people just to get the tribal knowledge that I talked about, right? To get that in, right? Where we have conversations, right? and then we with with experts experts, which is not written or documented anywhere, translate that into
[24:00] speech to text, right? Document that and and and feed it in, right? So, that and and and that is how your ontology is getting generated. But you have to have a human in the loop design, which comes in, right? And and is again able to
[24:13] validate it from an overall business sense perspective. It is not a technical exercise and should not be treated as such, right? Then, of course, change management, right? Of course, you know,
[24:25] for for people from from that perspective, okay, your agent doesn't work, they fire someone asks two questions, okay, did not give the right answer, so it's a problem, right? People have to understand it's a it's a gradual process and it's going to get better.
[24:37] So, that has to be handled with sensitivity and utmost care. Then you look at context curation. If you if you really look at context curation, it's not it's not one and done. Because your context is stale. Stale contexts, stale answers, right?
[24:51] So, you'll have to look at how to make the context, you know, more and more relevant, and you have to go on refreshing the context from time to time and the and the and that can be done on a weekly basis. I mean we've set set schedules for weeks, I mean for for
[25:03] weeks, but there are there are certain organizations who, you know, who want something daily, they want things scheduled ad hoc, etc. So, we'll have to cater to all of those. Evaluation, right? Normal out-of-the-box testing
[25:16] evaluation eval frameworks, etc. Especially for non-deterministic agent outputs, right? They will not work, right? You'll have to create eval frameworks with thresholds, etc. where you look at the agent more objectively,
[25:29] look at, you know, what's it trying to maximize or minimize? Is it hitting the thresholds or not? Or if it's below, what is ailing it, right? So, those things you'll have to document it and then of course there are feedbacks that you need to take, but the custom
[25:41] evaluation frameworks are are very important. So, while, you know, a lot of this and most of this can be done on Databricks, but, you know, there are certain cases where we might have to look a little bit outside here and
[25:53] there. >> Yeah. >> So, that's These were the challenges and, you know, most of them we've got our arms around it, but it's been a learning process. >> Awesome. Thank you.
[26:05] Um the next one we want to talk about hyper-reality, right? We are seeing it become reality on a day-to-day basis, right? It's not a futuristic vision,
[26:17] right? We we wanted to share our vision of how this is going in CPG industry and uh beyond, right? Right? Where are we headed in the next 12 months?
[26:29] A a big thing that we are seeing in the industry is using these agents, people have started agentifying the business process itself, right? So, analyze your current processes, right? System systematically
[26:43] mapping your existing business process workflows, identify high-impact opportunities for agent automation. To give you a few examples, um companies we're working at, for example, trade promo optimization.
[26:56] Right? Uh, trade promo optimization obviously is a pretty involved workflow a process. That's what we are agentifying quite quite a bit, right? To reduce uh the manual bits of that uh business
[27:10] process. So, is anywhere in your business pro- process there is manual work involved, how do you use agents to reduce that piece, basically? Right? And we have developed a proprietary solution called Rapid. Basically, what it does is
[27:24] you can put in any process in that solution and it will it will create the agents that are required for that particular process. Right? Within a matter of minutes, if not uh hours,
[27:36] right? So, your agents it designs the agents, it helps you configure those agents, it helps you give instructions to the agent, and it actually spits out the code for the agent as well. Right? So, using that uh a bunch of our
[27:48] customers are agentifying their business processes pretty uh pretty quickly. Right? One of them is your autonomous demand sensing, right? Agents that continuously monitor predict demand
[28:00] signals and triggering uh supply chain adjustments automatically. Another example is uh self-healing data pipelines. In fact, Databricks also is uh releasing uh self-healing data pipelines as part of their
[28:13] out-of-the-box uh feature. Right? And then con- conversational uh BI at scale. Obviously, expanding Genie-powered natural language uh access. So, Genie basically helps you answer questions like
[28:26] "What happened? What were my sale- sales last week? What were my sales last week compared to last year? Uh, which regions were doing better? Which regions were not doing better?" And things like that. But we have developed extensions to Genie where we
[28:39] can get into the why questions, right? The root cause analysis, etc. And what should I do about it type of questions as well. Right? So, um I I and deploying those type of agents at
[28:52] scale. And as part of your agent defining business processes, one of the most important things that organizations are doing or are going to do is multi-agent orchestration. Right? For complex cross-functional business
[29:04] processes especially. I mean, traditional business processes have been pretty siloed. And collaborating across functions has been a human's job if if it if the
[29:16] humans have have that kind of time, right? But now with agentic orchestration, that cross-functional you know, collaboration becomes a reality.
[29:28] And then finally, the federated context graph. Be it across business units, be it across geographies, be it across clouds, be it across platforms, you know, federating that context graph,
[29:43] uh abstracting it and then federating it across all of them uh is is what we see as uh you know, uh the the market going towards. Okay. Thank you everyone for your time.
[29:55] Appreciate it. >> Thank you. >> Have a good day.

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.