From Data to Decisions: Powering gen/Agentic AI with Capgemini & MongoDB

0:15Hi, everyone, and welcome to the Mongo DB podcast. I'm Apoorva, a staff AI Developer Advocate on the Developer Relations team here at Mongo DB. And in today's episode, we are talking about powering agentic AI applications with Cap Gemini and Mongo DB. I'm joined by Vinay Mukherjee, who's a Cloud Coe leader at Cap Gemini and Fareed Mohammed, an Advisory Solutions Architect at Mongo DB. Welcome to the podcast, Vinay and Fareed. Thank you. Thank you for having us. Yeah, I'm super excited to talk more about how Mongo DB and Cap Gemini are working together to help enterprises build modern applications with you here today. OK, how about we start with a round of intro? So tell us a little bit about what you all have been up to at Mongo DB and Cap Gemini.

1:05Maybe Vinay, we can start with you first. For sure. Yeah, this is Vinay Makanji. I'm part of the Cloud Center of excellence at Camp Gemini, supporting multi cloud Mongo DB in all all of the multiple industries like the mass, the CPR is the energy and the telecom. I've been here in the in IT for about 23 years, a member of Fortune 500 solutions team, Forbes Technology, Forbes Technology Council member, advisory leadership role. I've spent a lot of time in solutioning working with Mongo DB and Caption Night together almost for the last five years. Awesome Farid.

1:53Yeah, my name is Farid Mohammad. I have been with the Mongo DB for the last three years, so I'm based out of San Francisco Bay Area. So with over 2 decades of industry experience, right in my current role here as an advisory, I work with the partner ecosystem across the industries in the Americas region, where I support the partners to lead the enterprise scale, data modernization and a transformation initiatives, which in turn help our giant customers to unlock the strategic value from modern and data-driven kind of cloud native architectures.

2:31So I work extensively with all the global partners, including tech services, hyperscalers as well as Isvs to build the MongoDB Center of Excellence through certifications, through architect clinics, building the industry solution blueprints and also reference architectures. And I also help these partners with the solution architectures for large transformation deals to scale together and also improve our apartment channel business growth. Great. Well, welcome and let's get straight into it, shall we?

3:08So I'm sure the conversation with your customers has evolved quite a bit in the past two years, right? It started with curiosity, then there was this pressure to build and quickly ship prototypes. And then now we are finally at the point where we want to see value from all the investments that have been made into Jenny I. So can you tell me a little bit about what have been the biggest shifts you've seen across your customers? Where are you seeing Jenny? I actually moving the needle right now and what are the biggest gaps?

3:39Sure, I can start for it. So one of the back in 202223 and 24. We have been spending a lot of time in building the prototypes. We worked with Mongo DB, we worked with AWS, we worked with Google and also with Microsoft to build some prototypes. But this is now turning into reality, right? A lot of these prototypes have been shifting over in the last two years, especially in 2025. The the frontier models like the Open AI, the Anthropic, the Google DeepMind, Vertex AI, whatever you want to call, right? The latest solutions are shifting their focus towards reasoning, coding, safety, multi model understanding. The systems have evolved from just to support the art of completion of the code versus the support from code assistant to the project projects that we build. So AI product first is become our mindset now on a different parallel path from the generative AI, the software generative AI to the Agenic AI, it's turning into collaborations and that can plan and execute the tasks autonomously without even having a human spend a lot of time in the mundane tasks, right?

5:06That's where we're heading towards on the open AI, sorry, on the Agenic AI now considering the modernization of the tools, what we do is basically we learn the patterns from the legacy applications. We're not discounting anything on the legacy applications. We're learning new things from what the teams, the enterprises have built in the last 30 years and accelerating the processes in an enterprise, right. It's, it's easier said than done, but we are learning and learning and training the models in a way that it can accelerate the processes and the business sales.

5:42That's what we're heading towards. I'll stop here to see if any anything that you want to add for it. Yeah, sure, Vinay. So if I consider the journey of last three years, especially right after the release of ChatGPT towards the end of 2022, I would say we will be able to divide that into 3 phases. The Phase 1 during 2023 time frame is more experimentation phase. That is what we have seen with the customers where we are. We have seen a lot of proof of concepts around chat, intelligent chat bots, around copilots and also the isolated kind of pilots that the customers were trying to build and adapt to this new transformational technological shift with the innovative kind of budgets that they have, right.

6:36But we have seen good amount of progress and great results there, but it is disconnected with the enterprise systems. So that is one of the challenges we have seen. And then if we see the next phase starting 2024 for a year and year and half, what we have seen is more like now corporates are willing to operationalize this technology and trying to get the return on investments from the use cases that they are looking for. So that is where I would say the enterprise integration phase has come in.

7:12So even the organizations realized it is not just the models, but it's also the bigger challenge is the data that needs to support these models, the business proprietary context that needs to support this model and augment these models to improve the accuracies and reduce the hallucinations. So that is where the focus is to look into the existing data stacks, existing data architectures and how we can integrate and bring the data from existing different data sources wherever they are, whichever format it is.

7:48So that is where the architectures like retrieval, augmented generation, knowledge graphs have evolved, where we try to integrate and build, bring that business proprietary context to augment those LLMS. And then if I say last six months or so, right, I, we can take that as a phase three where now actually enterprises are trying to move from the experimentation to orchestrating the business outcomes. So that is where it is not only just generating the content and answering some of the questions. It is more on taking it to the industrial kind of use cases and try to see how can they leverage this technological transformation to automate some of the tasks as well in the form of agentic AI kind of world, right? Where if there are any reusable content or reusable components that they can automate and improve the efficiency of the overall processes is.

8:48So that is where we are in today. So if I want to summarize this, I would say the earlier generative AI was about generating the content and answering the questions, whereas the current agentic AI world is more about orchestration and driving the business outcomes. The adoption at the enterprise level has literally mature right? Embedding the AI into the core operations, whether it's sales operation, whether it's incident management or whether it's their own ITL processes or whatever it is, right. All the embedment of this AI into the core operations has changed significantly in the last three years. It is not anymore just responding to a current thing, right or just responding to a prompt anymore. It is, it is doing the auto healing, self healing, whatever you want to call.

9:41Plus there's also move from the proactive versions to the predictive version, right? That's what it's doing in today's world on the service management side. As an SRE leader also, it is important for us to focus on their operational area through through the AI solutions. Thank you so much. I think that was like a really solid summary of what's going on in this space right now. The one other thing I've seen as a developer advocate working with a lot of developers is this conversation around like we talk, we don't talk about models in isolation anymore or just data just in. We don't talk about any of these parts of the stack in isolation anymore.

10:24We are talking about Gen. AI. The AI stack has an ecosystem, which is something that I hadn't seen as much before Gen. AI. Like, there's no single vendor that can accomplish everything that needs to happen to build robust agentic AI systems, right? So what, according to you, does a truly effective technology partnership look like in this area? And on that note, can you also expand a little bit on how Mongo DB and Cap Gemini are building your unique partnership and what it means for organizations to use it to build Gen. AI applications?

11:01The generative AI applications build, I'll come to that in a bit, but very briefly wanted to talk about the partnerships, right. This partnership has not started today or two years back or three years back. It's been there for years. We've spent a lot of time together understanding what's needed for the customer. We have evolved from from the way the application integrate into the ecosystem of the of an enterprise versus how the generative AI applications will be built in the future state.

11:32So there's a lot of lot of study, there's a lot of analysis, there's a lot of research that went into before training the AI model and the generative AI. I would say I'll bank on the point that Fareed was making in terms of the RAG architecture, the vectorization that Mongo DB light platforms provide. But essentially what I'm referring to is all of these is driven by the platforms or it's not one vendor. To your point, yes, it's not a single vendor that's going to do all of it.

12:08Like take an example of a connected fleet. It's a it's a very good solution that Google and Mongo together came up with. And Cap Gemini uses it in majority of our retail space or in manufacturing space, right, where you want to track your trucks moving across the country, right. So this solution in the past has been very difficult with with 10 different applications integrating into within the enterprise, integrating into each other and sending the messages out.

12:41While today's generative applications are consolidated into, though it is not a single vendor, it is consolidated into multiple. Multiple things are consolidated into single solution that will drive the business outcomes faster. So the generative. Excuse me, the generative AI applications that that we are referring to has completely changed based on the technologies that we adopt into the system, right? One, the LLM, 2 the the rag architecture behind it, three of the models that we want to train, 4 the native solutions from the hypervisors, 5 the vectorizations provided by Mongo DB. All of them integrate together to provide a unifying solution for us to be able to generate the outputs that are required for the business.

13:30Yeah, absolutely. And to your point, right, Apurva, as you were mentioning like what we have seen, the companies that succeed in this a transformation, right, they consider this AI as a system rather than just a model, right? That is where they are trying to integrate and bring the systems together. So when we look into this generative AI and agentic AI, kind of a technological transformation, it is not just technical transformation, right? Always the companies like Mongo DB, they can bring that unified kind of intelligent data platforms with efficiently managing different structures of data, different formats of data and also provide real time kind of analytics and also vector search capabilities, so on and so forth, in addition to having the better scale, agility and resilience.

14:25But when it comes to the enterprise level success, right, that is where the enterprises are looking at having those industry specific domain based kind of reference architectures. They're looking not only from the data orchestration, orchestration perspective, but also from the governance and security perspective as well. And also they are looking at those system integrations to come together to provide required kind of a data to these AI based use cases, right?

14:56So that is where the partnerships with Cap Gemini will come for a rescue, where Cap Gemini can bring in that kind of a domain based knowledge, right, that they have with these customers, enterprise customers working with them for years. And on top of that, their capability of executing the transformations, big transformations, right, that is the experience that they will bring in. So when it goes with that, right, because any of these transformations as I said, not just technical, it spans across people, processes as well as technology. So when we bring in the technological capabilities, Capgemini will bring in those people and processes based on their domain understanding and based on their transformation execution capabilities.

15:49And that is where this giant value proposition is what is helping our customers in a way that we can enable them to move their use cases from experimentation to the enterprise scale AI systems much faster and more importantly by reducing the risk to the customers, right. So that is the key thing there. So that is where we are helping out gently to all our customers. Today's generative AI or the future state is it's in his own it's own cruise control mode.

16:22I would say it'll have to evolve with the human envelope. Yes. Also, rather than just assisting, building the models, assisting the business, replacing the systems, we would see it is a joint venture between the the IT organizations and the AI, as in its own organization, right together, evolve in the future to build the better future and make it real. Makes sense. What would you say are some of the most impactful? I think you started touching upon this a little bit, but what are some of the most impactful use cases you'll have tackled with your joint partnership?

17:05Do you have any success stories to share with our audience? Absolutely, Fred, do you want to go first? Yeah, absolutely. So this will also go back to our partnerships, right, Not only with the tech services partners, but also with the hyperscalers and all. Now recently we worked on a tripartite kind of a unified solution approach for a island island gas industry in specific, where we jointly build a tripartite kind of a solution based reference architecture to support the critical equipment predictive maintenance use case. So if I want to just highlight the reference architecture, right?

17:47If you want to showcase, should I? Bring that up. Yeah, absolutely. So, yeah, this is the reference architecture I was talking about, right, where all three parties came together, Mongo, DB, Capgemini and AWS. We have leveraged our seamless integrations with AWS to leverage their Bedrock and AIML model so on and so forth with all other services to bring the complete solution architecture here with the domain based knowledge coming from CAP Gemini side to solve any specifically this equipment health and performance monitoring use case and the predictive maintenance use case for the island gas industry. So if I want to give the perspective of this use case from the market side, right, if you look into the island gas industry, any unexpected downtime of any critical equipment as part of their production process is going to cost millions of dollars for them. So so far as Vinay has also touched upon, right, most of this predictive maintenance and all was happening as a reactive based kind of a mode based on the time based maintenance or usage based maintenance or a kind of rule based maintenance. So they were unable to

19:06anticipate when the equipment in the complete production process pipeline is going to break or take any kind of actions upfront to solve that right. And there is a recent study that has been submitted by Baker Hughes, which says at an average right, 80% of companies within this industry sector, Island gas industry sector, they have said at least they will see one unexpected kind of equipment failure with the critical equipment process in the pipeline within a year's kind of a time. And when it happens, it results into 250 thousand U.S. dollars per hour kind of an impact.

19:50So just imagine the scale because most of these equipments, right, they are managed by the third party manufacturers like Halliburtons of the world. So all these companies, they will procure those equipments, they set up their production pipeline process. And now how this solution is coming together to help them out is if I want to summarize this right, just imagine a imagine a agentic kind of a system where you identified which are my critical equipments that I need to monitor for in the existing pipeline. And then have some sensors associated with those equipments to track the telemetry and the usage of those equipments. And then once you store the telemetry data streaming that into Mongo DB based kind of platform, then you have all your legacy systems where you have your structural data related to those products, those equipments, those manuals, so on and so forth lying in your ecosystem. So that is where the benefit of the document model, right, where you can vectorize that data.

21:03So because most of these manuals are semi structured PDF kind of formats, you have some interviews done and capture the nodes from the past failures from those who have rectified it right that all lying there in your corpus. So just vectorize all the data and keep it handy in your one ecosystem in the Mongodbs platform. Now when you get this telemetry data, correlate the data with the respective past historical data and try to come up and identify and detect any kind of anomaly beforehand before the failure happens, right? So that is one of the automation that we can do. And then once you identify that anomaly, what you can do is you can now have the complete database of your historical failures and the resolutions done to those failures. So just do a contextual similarity based kind of a search leveraging the vector search capabilities and try to identify do you see any resonation between the existing anomaly versus the past failures that occurred and what is the recommended solution to fix that, right? And come up with that kind of a recommended solution approach to repair that as well.

22:15And then the other thing I can think of is just automate and have one agent to create the service ticket for this kind of an anomaly to look into and then pass it and assign it to a field engineer who can come and look into it proactively so that in that way they can avoid that complete failure and avoid the complete impact to the business, right. So that is one of the industry use cases where we jointly came together and build the solution blueprint.

22:44But I will leave it to you. Need to add more on this or any other use case as well. Yeah, I think you pretty much covered everything on the oil and gas at least. I just wanted to add 1 point to it. One is, you know, the solution started with just doing the RAG architecture and vectors, but it evolved again. It changed from how to automate this workflows. We started just automating the workflows and it's not enough for us. Automating the workflows was the first step, but we digitalized this thing.

23:15The digital trend was enabled the AIML hybrid models and then IoT related data streaming, right? And then we now went into the final phase of the AI where even before streaming the equipments are able. So the solution is able to identify what is going to be potential fault within the equipment before streaming and then that trains the data that is streamed out to the platforms, right. It becomes easier, faster and quicker to identify the solution and the turn around which will increase the productivity for sure.

23:58Yeah, I really like that we're kind of bridging the gap between like physical and software systems with this architecture. Like I know for a long time like IoT was it's own like large focus area, but I'm really liking seeing that gap between physical and software systems being bridged with this architecture. Very cool, thanks for sharing that. Can you tell us a little bit about where you're seeing the Mongo DB Cap Gemini partnership headed, you know, in the coming years?

24:29Like what are your short term plans, long term plans for this partnership? To touch upon that, we have been partners for years now in the In the future state, we'll continue to the partner the same way that we did for the last few years. But what we will additionally do apart from the database mass migrations or from the data warehouse or whatever it is, right? One of the recent examples that I will give you, my customer is also a Fortune 500 is a heavy MongoDB user is now migrating to cloud, is building a lake house architecture. When they're building a lake house architecture, whether it's Google's Big Query or Azure Fabric or Azure Data Warehouse, whatever it is, right, they're expecting to host the Mongo DB within their lake house architecture, right? They're not seeing the lake house and the transactional DB separate.

25:31They want to host the MongoDB as part of their lake house or the lake Big Lake or whatever you want to call for that specifically, right? Because the, the streaming, the, the, the platform as a lake host platform, they're going to build the AI analytics, right? And they're going to build as an example, let's say in the insights and data, they want to do any unified data management or a data estate modernization. They want to keep it simpler, binary objects, object oriented programming and then build search on top of it, build industrialized AI use cases on top of it and integrate vertically into the applications that are sitting in the call centre or sitting in the in the user interface, right?

26:19So they want to build that very closely. So what we want a solution tomorrow is not just the mass migrations, right? We don't want to stop with it. We will migrate the data. We'll modernize the data, yes, but we'll host the data to work with your lake house and to work with your applications and give you the insights real time, right? That's where we're heading towards. That's one of the major transformation that I see in the partnership that we have to work together and we've already started on it as we speak.

26:50We have at least three or four programs that are doing that work. Second, one of the biggest, biggest industry use case I would say is the mainframe or DB400 or AS400 modernization, right? Mainframe modernization has also been very converting the ASCII to epsidic to ASCII has been very difficult. Mongo comes with some out-of-the-box solutions that we could use and we are heading towards the mainframe data modernization into into MongoDB. That's another bigger use case that we want to partner on and we see a lot of opportunities there. We will grow together including the hypervisors, right, hypervisors and as as Cap Gemini and Mongo DB together will grow. The last use case that we are heading towards is the digital manufacturing and the industrial IoT, right. We've spoken about that briefly in the oil and gas piece, but I wanted to go outside the oil and gas as well where the factory of the future or connected vehicle, fleet management, fleeting sites, all of these or the IoT use cases that we can partner and solution for our customers.

28:10So we are seeing a positive outcome in the next 5 years at least that there is an opportunity for us to collaborate and help the customers build their solutions. Very exciting. I'm very excited to see how the IoT collaborations pan out. Yeah. Just wanted to add to that Apurva. So for Cap Gemini, right, we are not just a choice of a database anymore. So what I would say is we are Co investing and Co innovating with Cap Gemini and we are becoming that strategic enabler for them to win and accelerate the outcomes with protecting their margins and also scale into multi year kind of programs, transformation programs much faster, right. In some cases, we have seen almost like 30 to 50% kind of a improvement as well.

29:03So that is where this partnership is going to head in the coming years. So I want to summarize it as technology builds the capability, whereas partnerships build the outcomes. I really, I think that's a great note for us to close out on. I'm going to remember that one. But before we close out, do you have anything else you'd like to share with our audience? I would say yeah, please, please reach out to us if there are any questions, we're happy to help.

29:37We want to work on the latest and the great greatest use cases right together as as partners with hypervisors, Mongo DB, Cap, Gemini as an SI, all of all three join the hands together. But feel free to reach out to us or to this group and we are able to help. There are a lot of things that are happening. AI is changing, AI is evolving. AI is is encouraging to everybody as well. See it beyond the lens of just supporting you or reshaping the work, but it is also going to be a very big initiative program across all the organizations. Yeah, just just wanted, want to add one more thing, right? AI initiatives will stall at a pilot stage if you don't have a strong data strategy to support that, right? So that is where there has to be an alignment between data ecosystem architectures and also the people. So that all needs to come together to make these AI systems successful.

30:44So very kind of a thoughtful discussion. Thank you for that. Yeah, no, this is great. I had AI, didn't realize how 30 minutes went by. So thank you so much for being here and sharing all your insights with us. Thank you. Thank you so much. So much, yeah.