Sponsored by: EXL | From Dilemma to Decision-Ready: Turning Data into Actionable Intelligence

Summary

EXLdata.ai is an agentic-native platform built on Databricks technologies including Delta Lake and Unity Catalog, using 82 coordinated AI agents to automate data ingestion, transformation, governance, and lineage tracing.
EXL brings over 500 Databricks-certified team members and 400 trained practitioners, bridging foundational data work with agentic AI to help enterprises modernize legacy systems and become AI-ready.
This video covers EXL's approach to reconstructing data lineage from source to insight, integrating technical and business metadata, enabling continuous proactive anomaly detection for data quality, and activating unstructured data through semantic enrichment.

Chapters

FAQs

What is EXLdata.ai and how does it work with Databricks?

EXLdata.ai is an agentic-native platform that sits on top of Databricks technologies including Delta Lake and Unity Catalog, using a network of 82 coordinated AI agents to automate data lineage, governance, ingestion, and transformation. It is designed to be source- and target-agnostic, meaning it can connect to legacy systems including on-premises COBOL environments and route data into the Databricks Data and AI platform.

How does EXLdata.ai handle data lineage tracing?

EXLdata.ai deploys dedicated data lineage agents that trace data from its original source all the way through to its final target, providing full traceability for compliance and governance. This lineage is integrated directly with Delta Lake and Unity Catalog on Databricks, making the provenance of every data asset visible within the platform.

What is knowledge graphing in EXLdata.ai and what problem does it solve?

Knowledge graphing is described in this video as one of EXLdata.ai's newest capabilities, focused on activating unstructured data through semantic enrichment and ML-based classification. It addresses the challenge of unstructured data complexities that undermine trust in analytics and prevent enterprises from using all of their data for AI workloads.

How does EXLdata.ai support enterprises across multiple cloud environments?

EXLdata.ai supports deployments on Azure, AWS, and GCP, making it suitable for organizations with multi-cloud data estates. The platform's agnostic architecture means the same agentic capabilities for lineage, governance, and data quality can be applied regardless of which cloud the underlying data infrastructure runs on.

Full transcript

[00:09] Welcome everyone. Thank you so much for coming to this session, especially with it being the end of the day. End of the the very last session of the day. So, you we our booth is right over there. We have pens, we have gifts. Join
[00:23] come over there after the the uh presentation and we'll go through a demo and give you a pen. So, we are going to talk today about
[00:35] turning data into actionable intelligence with EXL data.ai. Now, EXL is an amazing organization, not because I work there, but because it I see what they do across the globe and
[00:48] how we're helping our clients deal with the foundation of data and the problems that exist today in order to modernize, become AI ready. Um so, we work with we serve our clients
[01:00] across the globe. We And as you can see here, we have 50 plus global locations. We have 127 global to uh global accounts uh clients and
[01:13] we are very um involved in the um uh setup of our clients. So, you see we have 800 global client clients that are
[01:25] existing today. And so, um I won't go through the financials as you can see on that slide. What we're going to talk about today is our partnership, most importantly, with Databricks. So, Databricks has been a
[01:38] great partner of ours as we start as we started to work on this concept of EXL data.ai. We have them as our launch partner at the time when we launched this back in
[01:50] October 6th. And we do get a lot of questions when we start talking about the deep diving of this data and how this works and the technology associated to it is, you know, how does Databricks and EXL Data
[02:05] Data Data Data Data work together? And it's more or less, I think Databricks can do all of this. But in reality, there are foundational work efforts that need to take place that we work closely
[02:19] with our clients to bring that data forward into Databricks. So, we have 500 certified uh team members within our organization that are certified in Databricks. We have 400
[02:31] trained practitioners. So, we bring forward a lot of knowledge in the Databricks space as we're implementing our Agentic AI solution. So, next up here is we're just uh
[02:43] talking about a bit of an overview here. EXL Data Data is Agentic native uh a platform that's agnostic to solutions. What we mean by that is that the source is agnostic to
[02:57] the capabilities and the target is agnostic to the capabilities. So, we have um the opportunity to really go into our clients. It doesn't matter what that language may be. We have clients
[03:10] that are still on prem today dealing with COBOL language, dealing with assembly, and trying to get off of those mainframe systems and bringing it forward to the cloud and to be in the uh AI readiness space. So, we really focus
[03:25] on the migration of that to help our organization modernize. So, we migrate and modernize our capabilities in the agents, analyze the code, analyze the data, and really bring forth the logic
[03:38] and give optimization in that space. And then, of course, in the end, we focus on quality and governance. We also have unstructured data. So, um it's it's truly powerful. And all the modules work
[03:52] in a multi-orchestrated way in that platform. So, from here, I'll go through the EXL data.ai framework. So,
[04:05] when we came together as a team and the co-builders of this of this cap- capability, we thought about it in a perspective of the full foundation of moving data.
[04:17] So, when you think about it, a lot of times we'll start in a migration or intelligence of migration perspective. Cuz we want to look at the legacy code. We want to see and peel back the onion and actually go deep in
[04:31] that logic that has been built and transform that logic into a um uh a language that can be accepted within the platforms that exist today. More importantly, what we do uncover as well
[04:44] as optimization in those jobs. Those we can peel back that onion and bring together 35, 40, 100 jobs that may be running and reduce those into three based on the analysis that's been running.
[04:56] So, that's where we get into the automated uh capability of engineering. And um what more importantly in the uh data uh migration space, we offer lineage. So, it's very interactive
[05:09] lineage. We think about it in a semantic layer. We talk about the the agents return information through the AI assist on really what they're seeing. What is that transformation logic? How is it
[05:22] moving through the uh source system all the way out to the target? And by the way, we have medallion architecture in this in the middle of this. So, we go through the bronze, the silver, and the gold layer before that's persisted out
[05:34] into a Databricks um uh platform. And then all around that as well, we bring data quality forward. So, we're looking at data quality from the source through the pipeline, through the
[05:46] lineage, and then out into the foundation of that. All along layered on top is the data quality and the data governance. We have the workbench, the command center that relates to the data
[05:58] quality that you interact. The agents are giving you recommendations on the quality and the health of the information. The agents are also actually suggesting data rules that may not exist, and that may in fact create a
[06:12] problem if if it's not applied. And then lastly, we you know, of course we talk about unstructured. So it doesn't matter structured, unstructured, we bring it all in together.
[06:24] And we have data observability. So through the pipelines, managing the pipelines, understanding those pipelines. In our governance space, we do bring in regulations and we also bring in the
[06:36] compliance space. We set our agents up supporting whatever the vertical may be at the time. And each agent supports whether it's health care, whether it's insurance. Those regulations sit inside
[06:49] those agents so that that's monitoring it out of the box. And then of course our clients could go ahead and offer additional rules, different regulations that we we may want to incorporate.
[07:03] And that's a very we bring that into the clients and we would do that. And then we have we've been getting a real lot of requests and a lot of insight on master data. So that 360 view
[07:17] of data. It's not like the traditional master data that's been in place in the past. This is really really bringing forward those core attributes that you want to monitor and make sure that as
[07:31] they're persisting out through your life cycle, whether it's downstream systems are returning it back to upstream, the agentic AI capability helps to manage that work effort through those pipelines. So, it's pretty powerful.
[07:50] Next up is just giving you a little bit of that path of travel. So, again, engineering, migration, data quality, MDM, reference data, operations, governance, and unstructured. All the
[08:03] underneath each one of these capabilities sits agents. In totality, there are 82 agents that we have Agentic AI agents that are running underneath this today.
[08:15] They are designed through the learnings of the build of our domain knowledge. We have a strong EXL has a strong 26 plus years of domain knowledge in many, many,
[08:27] many industries, and it helps us to really build out these agents with a flavor of LLMs as well as with understanding of the content through testing of data.
[08:40] So, more importantly, when we bring these to the client, it's out of the box, and then we start to work on modifying or customizing those agents as we bring them into the organization.
[08:54] I think the most powerful piece of this that we just added is knowledge graphing. Most importantly, starting with ontology. So, we have ontology graphing that's happening today, where it's defining the course of action of
[09:07] the data. Moving into the knowledge of understanding what does that mean? So, we're looking at it in a knowledge graph perspective, and we showed some of that in a demo right over there. You could come over
[09:19] and see a a live demo of it. But, what we showcase is we look at those sources. We look at those pipelines all through and the knowledge graph is growing. And then lastly, the context graph is where
[09:32] we start to see how it all comes together with a customer and a policy or a customer or a member and and a doctor and or physician. So, it all starts to grow out to tell the story of what that
[09:46] operation is. So, next up here is we're advancing this data management across the enterprise. There's a misconception in some cases,
[10:00] as you can see on the screen, is you know, we're just another data platform or warehouse, a point of automation tool, um limited to structured data,
[10:13] um consulting methodology, um and or playbooks. We're not This is Agentyc AI learning and there's a lot of misconception of what this capability is. It is truly a learning and a
[10:27] reasoning um a platform. So, it's taking the data, it's connecting the data, understanding how it behaves in an operational perspective, and and giving you that
[10:40] insight, key insights, overview, understanding, especially in our unstructured data. It could It transforms it into structure, but it goes further than that. It really truly comes back and
[10:52] gives reasoning for the data that you that it has ingested. So, for an example, we could take a lab report, if you will, and it could be 40 pages long and an X-ray and any other anomaly
[11:07] modality that would exist for an unstructured, and it brings it in and does create that structured data, but in the end it also provides the insight. What was in that document? What's the
[11:19] reasoning of that? And you can prompt the information and ask questions back and forth to the data through AI assist. So, that's again the most powerful piece. So, we we are uh and like I said, it's
[11:33] very I uh it's gigantic. It has an intelligence layer. And of course, we work in a lot of regulated industries. So, it it offers that explainability in
[11:46] the lineage. And also, too, uh we have that uh de-identification. So, that will also put uh masking on that.
[11:59] Here's a few case I'm going to move into a few case studies. So, modernizing healthcare with Databricks. So, here you see the challenge that we faced, legacy data. And we see this not
[12:11] only in the healthcare space, but we also see it in the banking space, uh a lot in the insurance space. You know, things are uh uh legacy data, legacy SQL, organically
[12:23] grown, no visibility to how it was built. So, we see a lot of that. So, as an example, this state of they struggled with the data warehouse. There was lots of performance issues, architectural
[12:35] complexity, and you know, the data was latent. And uh available availability was weak. So, and again, when you're thinking about that and if you hear from
[12:47] a lot of your clients, if you are in a frequency where your data is being refreshed only a certain point in time throughout the day, you're have a lot of latency in that. And in order to have the insights you need, if
[13:01] it's running over the night, you have a lot of risk. So, with that, this this particular scenario had similar problems. So, the solution was to implement Azure, and our platform sits
[13:14] on Azure. It also sits on AWS, and it also sits on GCP. So, we have covered all of the core uh spaces where you we can implement this um platform onto in
[13:27] inside a client's off uh uh uh platform. So, um they had Azure Azure Data Factory. We um recommended blob storage, uh Delta li-
[13:41] uh uh pipelines, Python, and Power BI. So, all of that was all inclusive in the solution that we offered and achieving scalable governor uh governed analytics.
[13:54] So, what we did is that's a simple transformation off of legacy data into a platform and warehouse that is more apt to growth and and and to grow within
[14:07] that and give you scalability. So, insights are the improved performance naturally, um near real-time scaling, which is another big factor, and reduce infrastructure cost.
[14:20] The decommissioning win that a lot of our clients get is really truly what comes out of this modernization of this platform because the all of these
[14:33] modules help to bring forward our clients off of that legacy and disparate uh capabilities that they have. So, improving healthcare data quality. We love healthcare.
[14:46] So, our data qua- I the out of all the modules, the the unstructured and the data quality is my favorite. In the data quality, you think about uh profiling, standard rules
[15:01] within the profiling capability. It goes beyond that. A Zectec AI goes well beyond that. It not only does that, but it also goes deep in understanding the data again in a reasoning perspective.
[15:13] They look at the the agents are learning from the rules and again, it's all within the the client's ecosystem. They're learning from those rules to expand on their knowledge. So now they're also
[15:26] actually looking at the data quality and how it's designed, but more importantly, they're reasoning with it and providing new information, whether it's a new data quality rule or that there's a a trend
[15:38] in the attributes, that there's you know, health one day, the health dips. You're getting all of that knowledge right in real time. So in this particular use case, we had a custom
[15:51] They had a custom SQL server. Very delayed data availability, which is normal when you have a you know, on-prem platforms are not very latent data.
[16:04] Architectural, you know, complexity and weak governance. So with that being said, our solution that we brought forward, again, was on Azure Databricks.
[16:16] And that lakehouse with Azure Data Factory, and we do a lot of work in that space as well. And bringing forward the blob storage and Unity catalog within Databricks and
[16:30] the Delta Live pipelines and Python and Power BI. So it was all connected in one string. And what I like to say is the impacts are, you know, real-time responses and really great wonderful
[16:42] things that are happening as you see here. But if I take a comparison of what we do as an organization and bring it together into Databricks, there are some key things that happen. So we help the
[16:55] client with the foundation of their data. We help them bring it forward so that they can use these capabilities. We tag the data for for you know, compliance. We tag the data
[17:08] for uh masking and such. You know, all of that protection that is required. But most importantly, when we pass it to Databricks, it's tagging automatically in that catalog, in the Unity catalog.
[17:22] Secondarily, Genie now has healthy information, so it can churn and churn and churn and put outputs that are you know, you have confidence in it because we're bringing that data in throughout
[17:34] the agentic process to ensure that that data is is is powerful. Um and then what has been recently launched is bring your own lineage. I had a demo today with that. It is amazing. So, we build the lineage within
[17:48] our pipelines. Our agents come back with the logic of that and they follows the path line. We have the transformation. It takes that natural language and bring it forward so it's understandable. We have the tech
[18:01] language, but you also have natural language that describes what that conversion uh transformation is happening in that pathway. And then all of that data, we have a pipeline connected to the platform, goes out into
[18:14] Databricks. And what happens is a Databricks visualizes that as well, but it also then in turn collects all the transformation that just happened. And provides an output of what that
[18:27] transformation was when it left the uh the lineage and went into Databricks. So, it's a very powerful and we're it's a new uh a launch on Databricks side and we incorporated that into our platform
[18:39] and uh it's a pretty powerful space. So, we when we deliver these capabilities to our clients, there's a lot of cost savings, lots of improvement in their processes, and it scale that
[18:53] brings them forward to scale even larger in the space of AI. So, lastly, our strategic partnership with um with the Databricks is very very strong.
[19:07] We we have a partner stewardship program. We're building partner of choice positioning. We have a clear structured partnership and go-to-market mo-
[19:19] motion and sales and people enablement. So, our partnership is committed to building strategy strategic partnerships. So, in that one is with our Databricks
[19:33] family. And we work very close with them. Our people are trained on Databricks, but more importantly, we've gone through quite a bit of a journey over the course of the year and have really truly perfected this
[19:45] capability that now is introducing the what's coming out in the market as well, which is that context layer.

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.