Sponsored by: Impetus Technologies | Powering Context-Aware Agentic AI at Scale

Summary

Impetus Technologies defines a "context gap" as the space between what AI models know and what makes a business unique, breaking it into four distinct gaps: the data gap, the semantic gap, the execution gap, and the trust gap.
This video explains how Impetus LeapLogic Suite automates modernization of legacy data estates—including ETLs and enterprise warehouses—with end-to-end lineage visibility, claiming a 43% acceleration in migration timelines compared to competing tools.
After modernizing data onto a unified platform, organizations can build a semantic intelligence layer and deploy production-ready, context-aware agents on Agent Bricks on the Databricks Data and AI platform with built-in governance and auditability.

Chapters

FAQs

What is the context gap in agentic AI and why does it matter?

The context gap is the space between what an intelligent AI model knows in general and what makes a specific business unique—its data, logic, and domain nuances. Without bridging this gap, AI agents cannot deliver the business value organizations expect because they lack the context needed to act, reason, and self-correct within a specific enterprise environment.

What are the four gaps Impetus identifies that prevent agents from delivering business value?

Impetus breaks the context gap into four components: the data gap (legacy silos preventing unified data foundations), the semantic gap (agents cannot understand data without a semantic layer even on modern platforms), the execution gap (scaling and orchestrating production agents), and the trust gap (ensuring auditability, governance, and drift detection).

How does LeapLogic Suite compare to competing migration tools?

This video states that LeapLogic achieves a 43% acceleration in migration timelines compared to competing tools, based on Impetus's experience with real-world modernization projects. The suite combines automated code conversion, schema mapping, and validation to reduce the manual engineering effort typically required for legacy ETL migrations.

Why is ontology-driven modernization important for AI agent accuracy?

Ontology-driven modernization ensures that when data is migrated to a modern platform it is also mapped with domain-specific business meaning and relationships, not just structural schemas. Without this semantic layer, AI agents pointing at gold-layer tables still lack the business context needed to interpret data correctly, leading to hallucinations or inaccurate outputs.

Full transcript

[00:09] Hi everyone, good morning and welcome to the session. My name is Sachnit and I lead the solutions architecture team at Empetus. Uh today I'll be talking about how we are powering context aware agent at scale and my focus would be why modernization should come first. I'll
[00:23] speak about what I mean by modernization and why we are saying that it should come first. First I'll talk about our mission. Our mission is that to engineer the context to make the agent AI to actually work.
[00:38] If you attended the keynote in the morning today, Ali spoke about the context. The LLM models, the agents, they are intelligent. They they can do anything for you. But if they don't understand your business, your context,
[00:51] they may not be able to deliver the value you are looking for. So we are bridging that gap. So, but what is that gap that we're talking about? We're talking about the context gap. The context gap is defined as a
[01:05] space between what those intelligent models know and what makes your business unique. Right? So, as I said, the models are intelligent, but if you don't tell them about your business, they don't
[01:18] really know all the nitty-g gritties of your business. They will not be able to deliver the value that you're looking for. And we are bridging that gap. Now, how are we bridging that gap? We've actually divided that into
[01:31] four different gaps, right? So context gap is not just one gap, right? We have to divide it based on where we are. Let's say if you are still stuck with legacy silos, your data is stuck in silos. You have uh legacy enterprise
[01:45] warehouses, ETLs and others. You're act data gap. The data gap is that you don't have your data modernized or unified on one single platform for the agents to use. That's the data gap. We are solving
[01:59] that through modernizing and I'll be focusing on modernize today. We also have a session tomorrow that will also focus on you know how we deliver the context which is you know building those uh agents and deploying them. Now second
[02:12] one is the semantic app. Now let's is let's assume that you are already on a unified platform right you have your data on datab bricks platform new platforms but can the agents read that?
[02:24] Can they understand it? You cannot just point your agents to the gold layer tables and all and make and you know assume that they will have all the context. You will have to deliver it to them and the semantic layer is what
[02:36] delivers it to the agent. So the second gap is the semantic gap where we engineer that semantic layer for our customers to make it available to the agents. The third one is the execution
[02:48] gap. You have built the agents maybe you are already experimenting with them but maybe you are not able to orchestrate them well you are not able to put feedback loops into it so which is which
[03:00] means you know how do you scale them how do you productionize them that's the execution gap and we are actually solving that through orchestrate our service software services on the orchestrate side and the last one is the
[03:12] govern which which overlaps right uh there should be security there should be governance and at At the same time, you should be able to audit. If you have deployed an agent which is going to take
[03:24] some decisions for you, you want to make sure that you are able to audit why an agent took a decision. What is the lineage to that? And also want to observe if there is a drift. You may have tested your agents, they were doing
[03:37] great, you deployed them, but how do you ensure 2 months later they are not hallucinating or there is no drift. This is where the govern govern layer comes in where we solve that trust gap. So we
[03:49] have divided this into data gap which we solve through modernization. Semantic gap where we engineer that semantic layer. Execution gap where we orchestrate those agents. Put an orchestration layer there. And the last
[04:02] one is govern where we have built solution accelerators to audit to observe all those agents that have been deployed. Now let's talk about what all solutions do we have. Now I'd
[04:14] like to introduce impetus leap AI family. Here whatever you see in the blue boxes are the solutions the accelerators we have built to solve all those gaps that I just spoke about. The first one is the impetus leap logic
[04:27] suit. I will be talking about that in more detail today. Then we have the impetus context fabric. This is the layer which actually builds the semantic layer solving the semantic gap that I spoke about. Then we have the prism
[04:40] observability. This is the govern layer you know where we make sure that we are have the right observability as well as governance in place. And the last one is impetus agent solutions where we have built agents agents to solve problems
[04:54] also agents to also you know cater to some of the repetitive business domain problems. I will be focusing on the leap logic suit today. Now let's talk about what leap logic
[05:06] suit is. I spoke about when I started why we why we are emphasizing that modernization should come first. Now depending on where you are in your journey right if you have already modernized you're struggling with
[05:18] semantic gap we have a solution for that if you already built the semantic layer you are struggling with govern orchestration we have that but what I'm focusing on is we we have seen a lot of
[05:30] customers who want to go agentic but they have not yet unified their data silos which means you may be running an Oracle a terara data an informatica maybe multiple BI tools a lot of business logic data is is out there and
[05:44] it's not just structured data. Your business may be handling a lot of unstructured data. For example, PDFs, documents, videos, meeting notes, a lot of that. How do we make sure that we
[05:56] make it available to the agent? So the first part is modernizing it, putting it into a unified platform like datab bricks. What on the left hand side you see are the legacy sources that I just spoke about. In the middle is the leap logic impet impetus leap logic suit. Now
[06:10] that that leap logic suit has different components or offerings as I would say the first one is assess second migrate third catalog then lineage and certify I'll speak about these what these
[06:22] individually do but the objective is that we start with an assessment of your legacy uh uh environment uh we create a lineage and dependency map out of it we modernize it using leap logic migrate
[06:35] which is taking you to a datab bricks uh modern platform the third one is creating catalog out of it. You have modernized the data but where is the metadata right? We need to create that metadata. This is where leap logic catalog helps. And then you need
[06:49] lineage, right? You want to understand what is coming from where, what column was transformed onto what. We maintain that using leap logic lineage. And finally, certify. This is the automated validation. You're bringing a lot of business logic and data from legacy
[07:03] platforms to data bricks. How do you know everything is correct? Right? This is where leap logic certify really helps. On the right hand side is the target where we modernize everything to data bricks. We create domain products out of it and then it is also augmented
[07:17] by impetus context fabric. Context fabric is is the is the family of solutions and accelerators we have built for the semantic and the ontology layer. As I said we'll have a session about semant context fabric tomorrow. I will speak more about leap logic suit.
[07:32] Now what does this leap logic suit in action actually mean? On the left hand side you will see a number of legacy ETLs EDW BI tools. Some of these may sound familiar to you because you may be using all this in your business. Now
[07:45] when you want to be agentic ready, right? How do you make sure how do you take it to the data bricks platform is where the modernization piece comes into picture. The leap logic suit comes into the picture and we follow a four-step
[07:58] methodology where we start with an assessment which is looking at all your legacies you know looking at the volume metrics the complexity the technical debt and then coming up with a modernization plan. Second is actual
[08:10] transformation. How do I convert an Informatica ETL to maybe a datab bricks lakeflow? How do I convert let's say a terod data or an Oracle business logic B tech scripts stored procedures to let's
[08:22] say run on data bricks platform or how do I convert your let's say Tableau dashboards or maybe any other BI tool that you're using to maybe something like AIBI or Genie on the datab bricks platform. You cannot do that. You cannot
[08:34] just throw an army of engineers to it and do it manually. You need tooling there. And this is where transformation step comes in where we use agentic to also transform this to the modern platform. The third one is validation
[08:46] which means if I have migrated your ETL jobs from Informatica to lakeflow how do I validate them? Again we are using agentic there we using a lot of automation there. And finally it's the operationalization which means I have
[08:58] modernized everything but how do I just run it on databicks platform. What about CI/CD uh processes? What about running those parallel environments validating it? So we take care of all that through
[09:10] this four-step methodology with our leap logic suit and we do that with up to 95% of automation. So if you are stuck on the first which is the data gap, you are still planning to modernize your
[09:22] legacies to databicks platform. This is the methodology that we have will which will really help you accelerate your modernization to the databris platform. Now while we talk about these
[09:34] modernizations right I spoke about B tech scripts I spoke about stored procedures but are are we just migrating the business logic as is for example I'll just migrate a a terterod data oracle stored procedure or terod data bx
[09:46] script to let's say spark code on data bricks no the new era of modernization means that while we I'm modernizing your business logic I should also keep in mind my ultimate goal my ultimate goal
[09:59] is also not just to modernize, have a lakehouse, but also to run agents over it, which means I need to plan for that from day one. And this is where ontologydriven
[10:11] migration comes into the picture. Which means if I'm let's say modernizing or consolidating your warehouses from legacies to databicks platform, should I just migrate everything as is? No, we
[10:23] should not. We should look at how do I divide this into data products, right? what kind of data products would be useful for your business and maybe the agents that you want to build right depending on the business uh the
[10:36] vertical that you are in what kind of ontologies should I follow should I follow a BFSI fibontology should I follow a manufacturing or a healthcare licensing ontology that's where the
[10:48] modernization piece really comes into picture when you think about not just modernizing the business logic but you think about that I need to have this platform ready for agents tomorrow. This is where ontology migration comes into
[11:01] the picture and we s follow that methodology through leap logic. The second one is the semantic intelligence foundation, right? I modernize your business logic, your ETLs, your warehouse. I have the gold
[11:13] layer ready now. But is it consumable by the agents? Can I just provide those two agents or do they need more? This is where also while we modernize building that semantic intelligence foundation comes into the picture and we do that.
[11:26] We do that using our accelerators as well as maybe the solutions that we have built or maybe if you are looking for building everything on data bricks native we use something called software services where we build everything for you on the datab bricks platform itself.
[11:40] The the third one is the context propagation right so how do we make sure that we are not bringing just the structure data for you but we need the complete context how does that propagate right how do I make sure that I keep it
[11:52] in sync so that while your business grows you don't have to redo the context layer again and again right and then finally we actually follow the agentic AI transformation which means that we would also be able to leave behind some
[12:04] of these components for the business as usual once I transform once I m migrate and modernize what what do you do if you have similar use cases? We can always help but we follow an agentic AI transformation which has all this in
[12:16] mind while we do the all this modernization. Now what kind of logical architecture are we looking at if you want to be agentic ready right we're talking about these layers right the first one is I
[12:28] I'll start from the bottom right which is you know discovering what you need where you are that's the first stage right we cannot just jump onto creating agents or creating a semantic layer we need to understand where you are so that we can plan that this is where you know
[12:42] discovery comes into the picture the second one is you know after while I modernize you right where am I going to build your metadata layer and how I'm going to enrich it the metadata is not just the tables and the and the column
[12:54] names right it is much more than that so I need to also discover that along with the technical metadata that I create for you on unity catalog how do I enrich and design it for you this is where you know broad business glossery the data product
[13:07] design you know ODPS ODCS schemas come into the picture because these actually define data contracts data contracts data products right now I have built the metadata layer which is technical metadata but then I'll also to enrich it
[13:20] to bring all that business data metadata. The fourth one is building actually building the semantic layer for you, right? That semantic model will actually have the knowledge graph as well for you. If an agent needs to
[13:32] answer a question for you, it needs the business glossery. Yes, it needs all the metadata and the access to the data. But it also needs a knowledge graph. This is where we build that semantic layer for the agents which has everything not just
[13:46] technical metadata but business glosseries uh the other metadata and the uh other information that uh that it needs. The fifth one is the knowledge core right which is you know now you have built everything right now fifth
[13:59] and sixth how do how do you expose how do you make it available to the agents right this is where you know we may we may build an MCP server for you we may expose all that meta data through MCPS so that it is not just exposed to the BI
[14:12] tools that you use maybe you are using AIB fine if you want to use let's say agents tomorrow you are building your agents deploying that to agent bricks do you build a new semantic layer for them no you can actually use the same
[14:24] semantic layer the same semantic core and expose it to your agents through maybe MCP is there so and the sixth one is you know governing after you have deployed everything how are we governing them how do you make sure sure that it's
[14:36] not hallucinating how do you make sure that you're checking for the drift in let's say 2 months time 3 months time and how does the feedback loop come into the picture so this is what a logical architecture looks like and we are delivering this logical architecture for
[14:49] our customers using our software services. Now, how do you what are view? So, what is our point of view when it comes to you know deploying production ready
[15:02] context aware agents to productions? I spoke about modernizing I spoke about creating a unified data platform creating a semantic layer over it and it's then it's all about you know uh building those agents. Uh we also have a
[15:16] solution called impetus context fabric which actually accelerates building those ontologies building that semantic core building that knowledge graph for you and having it ready deployable and
[15:28] accessible to the agents. We have a session tomorrow where we'll talk about the context fabric. But if you are looking for building it datab bricks native, you don't really have to go with the impetus suit of uh context fabric or
[15:42] the solutions. We can build it ground up from datab bricks native services itself. Agent bricks is what we are also using to deploy a lot lot of agents. Uh we are making sure a few things there. For example, while we do that, we are
[15:55] building all those agents from ground up where they understand your context so that they don't rift. We are making sure that we are testing all those cycles, you know, how it's is it reasoning, how is it validating and providing all those
[16:07] inputs. So, and also the governance part of it, who can use it, what can the agent answer and what can it not answer. So, we are actually taking care of all that in our uh methodology. just in case
[16:19] >> I have a few snapshots. We'll be happy to, you know, speak about this more at our booth if you want to look at deep dive demos. But also I have snapshots. I spoke about assessment. What does the assessment look like? This is a sample assessment of a warehouse like Terra
[16:32] data, right? It has it has volume metrics. We also look at you know what kind of technical debt may that may exist. >> Uh we may also look at you know complexity right hand side. If you have let's say 10,000 B tech scripts, how
[16:45] complex is your business logic to be modernized to the databicks platform? Right? Then we have this lineage graph which is which means you know if you are let's say I'm modernizing I will also keep a dependency uh map for all that
[16:58] legacy which will tell me that if I pick let's say a business use case I would understand that that business use case is actually using 10 tables. Those 10 tables are are being populated by let's
[17:10] say five ETL jobs. those five ETL jobs are actually using let's say two B tech scripts or three stored procedures and all so there's a complete lineage that leap logic is able to to keep then you know enriching the catalog from all the
[17:23] technical metadata that I have we will be able to enrich it with you know uh human readable information for example hey this table is doing this this column is supposed to do this and using all
[17:35] this information we are actually able to feed it for to our tool to tell us while I'm consoling validating let's say terod data oracles of the world to data bricks what kind of data product should exist right so this is this is a
[17:47] recommendation that the tool make obviously there's human in the loop your businessmemes our experts will s sit together and define what data products should be used or maybe improvise on that but the tool is able to suggest
[18:00] what kind of data products uh should be created right and then we create a pipeline which does modernization there are step which migrates all the data then we have a step which is transformation step which is you know migrating all the business logic to
[18:13] datab bricks platform and then having it validated. So this is an end to-end pipeline we create. Then we have automated validation. This snapshot actually shows how we automated in an automated fashion validate all the
[18:27] data and the business logic that we have migrated. So it runs it has uh it runs it on the legacy it runs it on the datab bricks platform and then compares and tells us whether the business logic that was uh uh uh u compassed into the uh
[18:42] Informatica ETL or maybe a data data B tech while I transform that to spark is it generating the same results or not is what this this is able to do in an automated fashion right now you know uh
[18:54] there are a lot of tools out there uh we had Macnite consulting conduct conduct a benchmark in January this year and they conducted a benchmark you know comparing leap logic against blade brbridge and what we were able to find was that what
[19:07] they were able to find was that you know 48% better SQL conversion came in for the same scope of work same business logic same complexity through leap logic at the same time there were 94% less
[19:20] manual efforts to convert that scope of work to data bricks using leap logic versus lake bridge and then you know overall if you use uh let's say leap logic versus lake bridge for such a modernization for an overaller timeline
[19:34] perspective it leap logic actually accelerates by 43%. And the other critical gap that they found was that you know if you have you are you are let's say modernizing your ecosystem which is BI tools orchestration
[19:46] everything blade bridge is not able to do that but leap logic is able to do that and then you know uh uh the gaps uh across SS phase across migrate across validate they have all documented that
[19:59] we'll be happy to show you more details how that was was conducted uh and also share the report with you if you will be interested more on that. So lastly concluding on what are the best practices right first start with the AI
[20:12] readiness assessment see where you are do you have the data gap do you have the semantic gap or you do you have the trust gap or do you have the orchestrate gap right first we need to understand where you are and then plan things from
[20:24] there uh modernize with ontology don't just look at modernizing the business logic look at how you will build the semantic layer and the ontologies validate before you go live this is very important and an automated validation
[20:37] like legacies were designed differently. While you move to data bricks, make sure you're validating all before you modernizing. Fourth, see how you can expose all that semantic layer, all that metadata to an agent as well from day
[20:49] one rather than thinking of it 6 months later. Then uh finally there has to be a governance layer so that you are able to audit and understand what those agents are doing. And if you're ready, you want
[21:02] to do an AI read assessment, we are there to help. You can book a two hours free assessment. We would assess and let you know where you are and what are the next steps.
[21:14] And I'll conclude by saying that we we have a session tomorrow uh which will talk about the context fabric I spoke about today. We'll talk about you know why agents fail and how we are making sure that the context is made available
[21:27] to those agents. We also have a booth booth number 400. Please stop by. We'll be happy to deep dive into more details. I have shared today the leap logic suit family or maybe discuss you know where you are in your journey and how we can
[21:39] help. I thank you for your time and enjoy the rest of the v guys. Thank you.

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.