Real-Life AI Use Cases on Cisco Infrastructure

I everyone, and thanks for joining today's session, our expert Expert insight series. This is real life AI use cases on Cisco infrastructure.00:00My name is Mark mcbrede and I'll be your host for today. Joining us today, we've got Joe Priyanka and Riley, who's gonna take us through today's session, covering topics on AI use cases across the industry, as well as covering some demos along the way.00:20Joe's the CSS engineer focused on data center compute container infrastructure as code and hybrid cloud technologies. Joe brings the history of using and deploying Cisco core edge and cloud infrastructure into production environments for over 20 years as both in engineer.00:36Here and as a manager.00:56Brianca is also a customer success specialist with over ten years of experience at Cisco focused on helping customers maximize the value of their technology investments and achieve business outcomes. She currently holds a CCIE and routing and switching and specializes in working with Intersight to help organizations.00:58Optimize and modernize their infrastructure.01:18She's passionate about emerging technologies and she's currently driving initiatives and AI driven solutions and customer experience. And Riley has been working as the CSS here at Cisco for over five years in the DC compute team where he's been focusing on inter site and Intersight workload optimizer.01:21Prior to that Riley was working in the data center for NetApp for four years doing everything from asset management to sysadmin level work, and this is where he found a passion for data center technologies. So here's an agenda for today.01:41We'll take a few minutes to do a previous session recap to get everybody up to speed on some of the topics that we've been covering in previous sessions. We'll also go over some of the trends and what are some of the directions that AI is going in.02:00We'll also cover some AI use cases that we're seeing across the industry and throughout the session, we'll also have a, a number of demos to show you how AI is being used. And then at the time we'll have at the end we'll have some time for, for Q and A in case you have questions, and we'll also guide you where you can take the next steps for.02:16Upcoming sessions.02:36So throughout today's session, we will be using Slido, so you can use the QR code to join@Slido.com and you can use the passcode aI-use case to join us with our, with our Slido polls. So we'll go ahead and launch our our 1st question, and if you're using the.02:40Webex application, you can also launch the Slido poll within directly from the Webex app itself.03:00So our 1st question is, where are you joining from? She could take a second just to let us know where you're joining from, myself and from Texas. We have some folks on the call that are also from, from Raleigh North Carolina.03:09See votes coming in, so somehow from atlanta, from Toronto, Appreciate everyone that is joining our polls and participating with us.03:31So we see a healthy number of participants from from all over. So we'll have a number of polls throughout the session. We'll continue using Slido for those. So thanks for everyone that's participating. Alright, and with that, we'll.03:47Get our session started and I will pass the ball over to you Joe.04:07Excellent. Thank you very much, Mark. Let me get my presentation ready here and what I'm gonna try to do is help our attendees follow along by using a pointer here so you can.04:13See where, where my focus is maybe at any point discussing some of these slides. So as Mark said, we're gonna recap our AI journey so far. This is where we have come from this.04:31Is where we are at today. As you can see, we came through a process of going over an AI portfolio overview as well as looking at customer maturity models. And after that, we went through several deep dives, went through one on data center networking for.04:47AI workloads. We also did one for data center compute for AI workloads. Those two are full of information, certainly recommend watching the recordings of those. We'll have that link.05:06Posted for you during this session today. And our current topic here is real life AI use cases that are running on Cisco infrastructure. This will be today's topic.05:23And we will have one more session in this series. We'll have one on agentic AI going from automation to autonomous networks. Again, this should all be recorded and viewable after the fact if you aren't able to join these live.05:38Now if we want to look at where we're at as far as AI ready data center, there are two possible ways that we can deploy that. What you've seen across the previous sessions that we presented is that Cisco offers flexible data center building blocks.05:57For your AI journey.06:16Now I wanna emphasize the word flexible here because I know you have existing infrastructure, you have established processes, and you're not looking to rip and replace everything. You have a couple of paths on the left here, we've got distributed AI infrastructure.06:22You choose your components such as next works Nexus switches for your networking needs, UCS servers for your compute and your preferred AI friendly stores. In that scenario, you can mix and match, you can build your own way.06:40And if we look over to the right, we're gonna see a vertically integrated solution here, got AI pods for fabric, AI, it's a turnkey solution where compute.06:59Network storage and the software all come together in a validated and ready to deploy package. And whether you want flexibility or maximum simplicity, Cisco has you covered. Now if you want to build your way with a modular approach.07:14We'll double click and look at those building blocks for networking, you have Cisco net nexus. And you already knew it, the nexus 9000 are running in XOS on top of the ACI, Nexus dashboard.07:34For AI workloads, we're talking about fabrics that can handle equipment and requirements of the GPU to GPU communications. There you're gonna talk about rocky V two ethernet. The networking session covered this in depth.07:49Again, I recommend watching the recording, and if you want networking networking is a service scenario where you don't have to manage that fabric by yourself, then hyper fabric will deliver that for you. So for compute, UCS again.08:07Likely familiar to you. These are modular, rack mounted DPU accelerated options, and these are managed through Cisco Intersight, which if you're not using that yet, is going to dramatically simplify your life cycle management of servers.08:23And these are available with validated designs, meaning that we've tested the configurations, documented any best practices, and you're not the 1st one that's going to be figuring out how to make it all work. And this matters when you're deploying something new.08:40Now for those who want everything integrated already, we have full stack systems, AI pods are validated hardware and software AI system. Now remember the compute, network storage.09:03It's gonna have openshift, Nvidia AI enterprise as part of that. Hyper fabric AI takes it even further where you're gonna have cloud managed integrated AI systems with silicon one.09:18Nvidia GPUs and fast storage, and these are all optimized out of the box for you. Now wrapping this up, our secure AI factory reference architecture, our AI defense for model security.09:36And hybrid mesh firewall for our network protection. So our big differentiator is security and observability that is built in from day one. Now we can take a look.09:53Get AI trends.10:13And innovation. As we see here, the state of the enterprise AI recent as 20 2053 points here that matter for planning conversations around AI. Enterprise AI spending hit 37 billion in 2025.10:15This represents a three fold jump from the previous year. Your organization is probably already spending on AI or your leadership is asking why you are not. But here's the reality check from our Cisco AI readiness report, only 13 % of organizations are truly AI ready.10:35That's roughly one of ten. Those space centers are outperforming their peers on every measure of AI value. Being AI ready is instrumental for success, which means you're in the room thinking we're behind when you're actually in the majority and have an opportunity.10:56The on premise and hybrid AI. These deployments are gaining traction. It matters for infrastructure teams. The initial wave of AI was send everything to the cloud. Now organizations are realizing data gravity is real.11:12And latency matters. Now we have compliance that's gonna require control. You're gonna be asked to run AI workloads on infrastructure that you manage, but keep in mind when you keep AI infrastructure close to your data.11:28Then you're gonna unlock capabilities that cloud only approaches simply can't match. Now here's what the research shows. Spawner alert, the best ROI is coming from back office automation.11:49At the moment, and what is clear are that the use cases need to be. Those are two times more likely to succeed. That's not experiment with AI, but let's solve this specific problem with AI.12:09Infrastructure readiness, now these are three times higher. As far as success rate goes, here's what I want you to notice. This is your domain, this is what you control. If the infrastructure isn't ready and the AI projects built on top of it will fail.12:28You're not just supporting AI initiatives, you're enabling or blocking them at this level. And when we look at workflow integration, three times more effective in that scenario.12:45AI that lives in a silo, delivers nothing, AI that integrates with services such as ServiceNow, with your common monitoring tools, with your existing processes, that's what delivers big value. You wanna take?13:00A iterative approach, these are four times better outcomes when you do, you want to deploy, measure, refine, not some big bang, but a continuous improvement model. As far as cross functional alignment.13:19These have four times improved results that can be in the areas of IT, business, AI practitioners, and operations all working together. And as I mentioned before, interestingly.13:38ROI return on investment. It's coming from back office automation at the moment, not some flashy customer facing AI. These are the current trends. Now if we look at the AI use cases across industries.13:57We have to ask ourselves what are our organizations actually doing with AI? So we're seeing knowledge based co pods and AI assistance. We see content and code generation, text images.14:14Videos and code being produced. We also see virtual agents chat bots that have specialized and specific domain functionalities. We also see visual computing things such as digital twins.14:30Video and analytics, imaging and diagnostics. And this is where one of our demos today will focus, keep an eye out for that. Language translation, these will be for real time multilingual communications and detection and prediction systems.14:47These are the generic use cases that are gonna be valid across multiple industries. But within each industry, there are specific use cases that drive the most value, and I'll speak to those later in the session. Now overall AI is transforming every.15:07Every industry. One of these probably represents yours, maybe multiple. We do have examples of AI across several industries we'll discuss today. Healthcare, e.g., financial services, manufacturing.15:27Retail, sports and entertainment. Each section has specific youth cases that drive value to it. And later we'll poll you to understand which industries are represented here in our attendees today.15:46And we'll dive into some of those examples. And here's what I want you to understand regardless of your industry, the infrastructure requirements are similar. Low latency inferences, data sovereignty, integration with existing systems, security, and observability.16:01No, I think we've got a poll coming up for you. Yep, and I'll turn the ball back over now for that. Thanks Joe, we'll go ahead and launch our next poll question. So it is who where do you expect to run most of your AI inference workload?16:21So the poll is up, you can take a second and answer, answer the question. Looks like most folks are in a still evaluating have not decided yet. We have some folks sharing that possibly a hybrid.16:41Using a mix of on prem and cloud looks like majority of are are still in a evaluation state and hopefully today's session will will help them.16:57Come to a deciding factor on where they can run some of these workloads. Right, thanks everyone for answering and I'll pass it back to you Joe. Excellent. Thank you Mark. For our audience, we want to look at this graph here, which is excellent.17:13As you'll see in the top left here, we've got our architecture split between private cloud, on prem, hybrid cloud, which a lot of customers find themselves in today, as well as edge AI. Now we have something planned for that that should come.17:33In a later session for us, but it's not ready yet. Now those scenarios are best known for enterprise wide model training, e.g., in the private cloud on premise. It's where you do your training and fine tuning.17:50And if you're going to do workloads that move between on prem and public cloud, obviously that's where you want to have your hybrid cloud model deployed. And if you need real time inferencing in remote or industrial sites, that's where you want to have your edge AI set up and optimized.18:06Now each have key advantages that we want to look at maximum security and data sovereignty and gravity as we discussed before, can be found in that private cloud on premise model. And if you deploy in the hybrid cloud.18:25Model, you'll have advantage of observable consistency across all of your environments, and for AGI we'll have ultra low latency and local data processing. Now those solutions here.18:42We'll see if you're gonna do the on premise private cloud. You'll be able to take advantage of Cisco AI factory within video that can combine models such as the UCSC 8805 and eight.18:59Or the X series with the Nvidia H100 or B 200 GPUs. The Cisco X series with Intersight, that's in a modular compute that allows you to add GPU nodes.19:14Can be utilized in that hybrid cloud model where you can add those nodes as needed and manage them via Intersight in the cloud. Now Cisco unified Edge, as I mentioned, we've got content being produced that's coming up for that. So we'll have more details around that later.19:29But for now you can know that it's a compact ruggedized three U platform integrated with Xeon processors and then video GPUs. Now to control those items, if you're doing the private cloud, you can do this.19:48Converged virtual appliance or the private virtual appliance if needed, which it often is in that scenario. Hybrid cloud model, you can do either of those appliances as well as the software as a service, Intersight in the cloud.20:02Controller, and for edge AI, that'll be the software as a service inter site and the cloud model and we can take a look at use case running in AI pod for our inferencing.20:19AI pods are the foundational and modular building blocks that you can use for AI infrastructure. It helps to think of it this way instead of separately procuring compute and storage.20:38And networking, then trying to figure out how to integrate all of these items and validate those. AI pods give you all of that together pre validated, and you still have the flexibility to choose your GPU options.20:54A lot of the configurations, but the integration work part is done for you. Today's focus is inference, and we'll be using models to deliver that value. But we have options for full life cycles as well.21:11Oh optimization in the context of AI machine learning is gonna refer to the process of refining and approving models for your systems so that they achieve better performance, accuracy, efficiency, and resource utilization.21:30This is gonna record or include techniques, things like fine tuning, where pre trained models parameters are adjusted for specific tax, and retrieval augmented generation or rag.21:49And that's gonna enhance models by integrating real time external data for context to where it outputs back to you. Now optimization aims to maximize outcomes and it's gonna do that while minimizing cost.22:05Latency and computational resources required. It's often gonna use tools like pytorch, HuggingFace transformers or video Nemo to streamline workflows for AI developers and IT ops.22:22You maybe familiar with some of those, and we can take a look at our industry use cases and see where they're accelerating or AI inventuring at the edge. One trend you'll increasingly deal with especially for industrial aligned use cases.22:37Is AI enriching at the edge? Data lives somewhere, maybe coming from sensors, cameras, machines, branches, points of sale, and the 1st wave of AI was sending everything to the cloud.22:57But that creates problems that you're gonna understand well, things like latency or real time use cases, cloud round trips sometimes don't work or don't provide the best experience. If you look at your bandwidth calls stream your video to the cloud can get expensive fast.23:12And things like data sovereignty or your regulatory requirements are gonna mandate keeping data logo. This creates a paradigm shift at some point for the centralized cost of model.23:32For the edge AI. Cisco's infrastructure, including our recently announced unified edge solution is gonna address this need exactly. Now for today's demos.23:47We can give you a kind of a sneak peek into the hardware behind that for the edge or inter tier deployments, we have UCS rack based.24:02The Servers within you unified edge option, but for dense compute supporting demanding AI like fine tuning, optimization, larger infancy workloads, we have our GPU dense models.24:20Now that have PCIe or NV link configurations. On our lab today, the demos that you're gonna see are gonna be running on AI pods for imaging and running on a UCSX server with GPU accelerators.24:39We want to get closer up and take a look at our AI pods, real hardware as we'll see in our demo. Running real configurations, which you've had to see isn't simulated, it's running on this particular infrastructure.24:59Oh, here we've got two intel, I 6526 wide processors with 512 gb of system memory. We're running five, 3.8 tb in the me drives for storage. We've got two NVIDIA L 40 S GPUs.25:16And these are tuned great for inference workloads. Now the software stack is a OpenShift and NVIDIA AI enterprise. This is the inference ready configuration. It's not some massive training infrastructure per se, but a practical starting point that delivers the results.25:35You may have something similar to this running in your environment already. Now we had folks from our direct team responsible for setting this up and you maybe able to connect with them at Cisco Live if you can catch them.25:55There you can learn about the tooling that they developed to help automate the initial setup of AI pods. Now with that being said, I think we're ready to jump into our 1st demo and I will turn the ball back over for that. Thank you. Thank you, Joe.26:10So, as you all explore how AI can transform your operations, you might often face the challenge of of identifying which AI use case can align closely with your business outcome and at the same time delivering tangible benefits in your network operations team.26:27So, so today I'm gonna walk you through two practical demos that could actually serve as an excellent starting point for your AI journey in data center.26:47So these two demos will be focused on, on prem offerings of Cisco that is called airports. Our 1st use case, as you can see here is anticipating network failures due to physical interaction and 2nd one is based on network operations. We also have a 3rd demo.26:57Demo which is a game which you'll enjoy when we play it for you. So that is focused on hybrid cloud deployment with AI pods in Gemini, which my friend Riley is gonna demonstrate for us.27:17So I'm gonna start with the 1st use case that we have here anticipating network failures due to physical interaction. So our goal with this use case is to identify if a network failure is, is due to the physical contact with the device.27:29It could be your chassis, it could be your server or basically any device in the data center. So now why would we want to identify that? Well, well, this could actually help us identify if any random hardware failure is unauthorized or authorized in.27:47Intentional or non intentional.28:07So there are so many times and then we have all experienced this that our port went down and we started to troubleshoot it and with hours and hours of troubleshooting, it turns out that it was some some undocumented hardware change or, you know, someone has has moved the cable without even knowing that they moved the cable.28:09So with this use case, we can actually detect that very early in our troubleshooting process and we don't have to spend hours and hours on issues like this. So now, do, could you pass the ball to me so that I can move the slides?28:29Surely take care of that real quick. Sorry about that. Thank you Joe. Okay, so, the talking about the setup, so we have a camera fee that is coming from our San Francisco office.28:49That is getting fed into a segmentation model, which is outlining a person and all the relevant devices. So in our case, these relevant devices are our devices in our data center. Then this, this feed is getting fed into BLM model, which is analyzing this feed and then converting into a.29:08A text. That, that text could be something like NO human contact, NO hardware interaction or, you know, person is exiting the frame, something like a person is removing a cable, person is modifying something a text like these, and that's text is getting logged into a log file.29:28Which are LLM LLM or MCP agent is reading that file every 10 s and once it identifies that there is something some some text like person is modifying something, then it's gonna trigger the MCP agent and that MCP agent is going to MCP server.29:47Now this inter site MCP servers is reading the inter site alarms. So why this is reading this? Because it want to identify if anything went wrong, if there is any alarm raised during the same time frame as this person made contact with that device.30:07And once, it identifies that it delivers the message that there is a port down due to a physical contact made with that device. So, we have a recorded demonstration because you know it's, it's hard to get access to the data center at the same time with this webinar. So I'm.30:24I'm gonna play that for you, but before that, just wanna go through what use case setups we have, what models are we using. So we are using, we are using VLM Coinci 8 billion parameters model.30:44For LLM and MCP we're using Gemma three and our video source which is an MPP four MP four file right now. This is from Meraki camera integration in our, our data center office.30:57Data center at San Francisco office, and then all these models are running on AI pods with two nvidia L 40 S as as Joe described our our lab environment. So with that, I'm gonna go ahead and share my screen.31:12And just give me a second so I hope you all can see my screen.31:31So before we go into, actual demonstration, just wanted to cover some capabilities of the solution. So the 1st thing is it's gonna unify operational context. What it means is it's gonna eliminate the silos between your physical security and IT operations. So you know both the teams can.31:46And I can see the same data, same output, and then same reports.32:06Then 2nd one is accelerate root cause analysis, which makes sense because you have your camera feed, you have the timestamp, so you know you can always go back to that timestamp and look if this if this is unauthorized, non authorized, if this was intentional, non intentional, so you can make decision.32:12Very early in your troubleshooting process.32:31And the 3rd one, AI powered self monitoring networks. So this is gonna distinguish if it's a spontaneous hardware failure or or if it's a human induced error. Then the 4th one is it's gonna generate the report that is gonna help you.32:35And simplify your compliance and investigations.32:55So I'm gonna play this video now. I'm gonna move ahead in a bit. We have already discussed the setup, so let's see this in action. So if you look here, before I play this video, we have four sections. So on top.32:58Left corner, this is the GPU utilization, which is gonna spike as you see the movements in data center here.33:19Then on top right corner is our data center, and then this is processed video from our SAM model segmentation model, so you might see a green border around a person or or some devices. And that this video is getting fed into the VLM model, and this VLM model.33:28That is analyzing that video and giving us text messages. So you'll see some text messages as I play this video, but now I've put on pause so you don't see anything.33:48And then on the bottom right corner is our MCP agent, which is gonna get triggered as soon as I detect something unusual happening. So I'm gonna play this video now. So as you see this GPU utilization is spiking because it's getting it's detecting.33:58A movement in the data center. So if you see this person, this is it's the this person is getting highlighted in green. And if we look at the bottom left corner, the VLM is starting to lock the actions, so it's saying something like NO security or.34:18Let me play this, safety issues. This person is just you know visually monitoring the devices. It's, it's continuously, you know, giving us messages like that. So it says person continue opening rack doors without engaging internal hardware.34:37Now LLM is continuously monitoring it and it is not seeing any unusual activity as unusual as in this person is not modifying anything. So it's not gonna trigger the, the MCP agent. So this is still nothing is detected this person.34:55Person remain in the same position without interacting with hardware and, just a visual inspection. So this person now leaves the data center and there is NO triggering of MCP agent.35:15Now, 2nd case scenario where another person comes to this data center and then starts to do some, starts in contact with the hardware devices, we still don't know if this person is actually triggering anything or modifying anything, but.35:29Since they start to, you know, modify something.35:49Will so this this says that visual inspection are actively manipulating internal hardware components. So as this person continues to do that, the MCP agent is gonna trigger it and then it is gonna go to MCP server inter site MCP server and monitor the alarms.35:54So in in this camera feed, this person actually didn't do anything wrong with, with the devices in our data center. So when we went to inter site MCP server checks, it, it checks detect the all alarms and there was nothing wrong and nothing raised in inter site. But since we.36:13We wanted to simulate everything for you, how the entire process would look like. We decided to bring one of the ports down in the same rack. So if you look here, this, the inter site will show you that we brought the port down and then there is.36:33And that adapter interface is down. So now we're gonna go back to this and since we know that we purposefully made this, this port down, it'll just detect once this person starts to modify something, this this MCP agent is gonna go to MCP.36:51Server and then MCP server is gonna detect that something is wrong because it will see that alarm and then it does the correlation that there was this there were changing from this person is changing something and then the port went down.37:11And then you could do your own analysis if this person was was authorized to do it, if this was intentional or by mistake, this person removed something that they were not supposed to. So with this use case, you could actually monitor your physical safety of devices, you if you want to re.37:30Reduce the time to troubleshoot to identify unknown physical errors made on data center. So this use case where will help you identify all of that.37:50So with that, I'm gonna pass it to Mark and he's gonna take us through a Slido question. Over to you Mark. Thanks Brianna. We'll go ahead and launch our next Slido question. So which AI use case do you believe to be the most relevant to your organization.38:01Right now. We'll give it a few seconds for folks to vote, whether if they're interested in document processing, network automation, possibly video analytics like Priyanka just demoed, maybe image analysis or maybe some other use case.38:21That looks like the majority of votes coming in around network automation and AI operations. And with that, we'll go ahead and pass it back to you Joe. Excellent, great. Thank you Priyanka. Wonderful demonstration.38:45And thank you Mark. I'm gonna take our audience on the journey now where we're gonna take a look at some use case examples across different industries.39:03And then we'll go through a couple of real life AI use cases. Now, before we do, let's take a look at the results here, the breadth of what we're seeing. In retail, e.g., we see point of sale transformation.39:20Customer traffic analytics, dry through optimization as possible, all using visual AI similar to what you just saw in Priyanka's demo. In the manufacturing industrial industry.39:40Things that are possible, can be product optimization, safety, predictive maintenance, having AI analyze sensor and camera data in order to prevent issues.39:56Now in the healthcare field we see we've got medical imaging analytics, diagnostic support, and here we've got AI assisting clinicians while keeping data on premise or compliance.40:15And the cross vertical edge, we've got edge in between servers, digital assistance, digital twins, and the pattern across all of these is that we have AI processing data close to where it's generated on infrastructure that you control.40:29Now let's show you what, what's we're working on in some of those specific areas. Here we're looking at retail use case. We're gonna go through several of these slides and if you would follow along, we want to focus on the outcomes here, the benefit.40:49To you for those in the drive through optimization, the outcomes we are gonna wanna see that you achieve the faster order processing, reduce wait times, increased order accuracy.41:05Higher customer satisfaction, improve efficiency. Now, those I'm sure are some outcomes that you might like to see for your retail organizations, specifically a drive through.41:22For that and to produce those outcomes, we're gonna want to overcome the challenge here of enhanced and optimizing customer experience in the drive through and for solution. Of course, we're gonna have several key items that we want to focus on as part of that solution.41:38We're gonna have things such as AI powered order prediction. We'll be able to take advantage of automated voice recognition. We'll have real time traffic management.42:01Be able to do dynamic menu display and the customer feedback integration into your organization at retail, you can imagine those are gonna be beneficial to your environment.42:16And produce some of those outcomes. Now over on the right, we're gonna see a kind of a mock up of the drive through. If you notice we've got the car, we've got the window I'm sure everybody here has experienced that. And the acrucialist there of what we would have deployed such as cameras.42:33Sensors to gather that data of what's happening in that drive through. Now we're gonna have Cisco comprehensive security on top of that, as well as full stack observability.42:51We'll have a partner listed here at the bottom where we can see maybe in that with us and have our software stack running on top of our AI pods, e.g..43:11And there we can incorporate Splunk intelligence. Meraki cameras, e.g., Cisco routing, and we'll have other items from Cisco that gonna be involved in that.43:30Again to produce the solutions to the challenge that provides the outcome to our customers there. Now looking at another similar slide here, this time in healthcare.43:45Again we don't need to read all of these, but for the outcomes in medical diagnosis there, we wanna see you achieve improved diagnostic accuracy, faster time to diagnosis, better.44:01Patient outcomes, reduced diagnostic costs, enhanced and efficiency, Ensure that's true of your organization, and we've identified the challenge there of a need for faster medical diagnosis of.44:19Diseases and for the solutions there, we want to provide solutions. They're gonna be things that automate tasks, detect abnormalities.44:36Personalized treatment. Now for that one, there could be an AI that can be used to personalize your treatment for patients based on their individual medical imaging data. And this can help ensure the patients receive the most effic effective treatment possible.44:53And decision support. We want to provide clinical decision support, then your AI can be used to provide support for your radiologists and other healthcare professionals, help them to make more informed decisions about patient care.45:11If we look at the scenario, we see that we've involved things such as x rays, CT scans, different devices. We've got our Cisco AI pods there. We've got our compute.45:30Our switching, our AI friendly data storage, utilizing utilizing nvidia, AI GPUs, and AI enterprise, got Red Hat OpenShift supporting that. We've got several partners listed there.45:48You maybe familiar with or utilizing some of those today and for our next use case, take a look at a manufacturing facility.46:07Where asset visibility and control, maybe top agenda items for you in the outcomes that would most benefit you items such as improved asset utilization.46:24Improve plant wide productivity, we want to reduce waste, of course, improve your worker safety, and reduce time to locate assets. We've identified the challenge of a lack of visibility.46:41Into your assets, their status, their condition, even their location, and we want to optimize those items in order for them to be better utilized.46:59And for the solutions, see that we can incorporate real time monitoring. This can be Cisco cyber vision and Meraki in the brand cameras. Centralized management, that can be Cisco DNA Center and Meraki dashboard.47:16We want to have proactive alerts. These come from Splunk and Cisco AI driven analytics, and we want to end up with operational efficiency, be Cisco's industrial asset vision.47:35Combined with Splunk's data analytics. And our infographic we see a representation there of manufacturing facility, person there looking for asset visibility and control and we've got our Cisco comprehensive security as always.47:51In our full stack of observability, see we're running our AI and machine learning software stack on top of our Cisco AI pods. We're doing our industrial switching.48:10Or industrial routing, and these are hardware models designed to operate in industrial environments that are hardened with specific features. Wi-Fi, of course, that's always useful.48:29As well as our meraki cameras and sensors. Now for our partners down at the bottom, you maybe familiar with several of those, probably already interacting with them. And this brings us to our next demo.48:44With that, I'll turn it back over to Priyanka and where we go. Thank you, Joe. Well, when looking at the last slide i'm glad that most of you are interested in network operations because our next use case demo is our network operations. So, if you.49:02Work in network operations, you know the reality. Tickets.49:22Syslog messages, alerts, and all flooding in constantly. Some are critical, but but most of them are noise. And when something happens, you have to go through all the logs trying to understand what went wrong, where it went wrong and how to fix it.49:27So now with AI, so now with AI, how can we make our lives easier, basically reduce these noises? So well with this use case, this is what we are trying to achieve. So this use case also came from automation and then utilizing the entire Cisco AI factory.49:46Including Splunk, your GPU enabled servers et cetera.50:06So what is happening is we are collecting events from our multiple domains and devices forwarding these events and alerts to Splunk and then some of the specific events are sent to webhook. And this is when webhook analyzes them and performs an automated response with.50:10The next steps attached to it with the help of LLM model, you know, so that you can just, just go ahead and quickly clear your tickets.50:30So, how are we doing it? So this is the architecture here. So on the left side, this is a webhook application, a custom webhook application which we made. Then on the right side is the architecture, so our our logs are getting into Splunk, then Splunk alerts to be.50:39Webhook services and then webhook is deciding which LLM to use. You could use your, your public LLM or other, something that is deployed on your AI pods.50:59And it is also integrated with your SMTP email so that you can send email notification and then also integrated with ServiceNow. So it's gonna take a call what actions you can take. And now I'm gonna show everything to you live. So just give me a second and I'm gonna share my.51:11My screen.51:31Okay, so this is our, our Cisco modeling lab. So this is where we have configured our nexus, which is our our servers and everything, and all the syslog messages from next.51:41Access switches are going to a Splunk. So our data center client one and client two are acting as servers connected to Nexus switches and to firewall and up to external connectivity. Our left side is NDFC domain and the right side is catalyst center domain. We're not gonna focus on Catalyst Center domain today. Just gonna focus.51:54As an NDFC.52:14So, and one more thing the since we wanted to generate some syslog messages, we actually created one secondary IP address in this VLAN on Nexus switches, which is the same as data center.52:15Data center IP address. Sorry, that server's IP address. So when I ping from this server to default gateway, it's gonna go to this Nexus switches and then it's got this Nexus switches is gonna generate identical the duplicate source IP address syslog, which is gonna go to Splunk.52:37And then to my webhook application. So this is how my webhook application looks.52:55This is the dashboard where we have list of LLM providers, alert type, all the webhook events. Now, then LLM providers, so we have the public one and the one that is hosted on our AI pods. You can add your own provider on top left, top right corner by.53:02Clicking on add provider.53:21And then we have alerts type. So these are all the alerts type that we configured on this webhook application. So link down peer BGP peer down, high CPU config change, and this is our duplicate source IP. So I'm just gonna click on edit and show you.53:25Call what we have configured for duplicate source IP.53:43So this is the description, you can use your LLM provider, you can choose the severity here, urgency, your LLM prompt, and then this is where you could enable this alert type. You can choose to use your LLM, choose to create your ServiceNow ticket and send email.53:47And now when you choose to create a ServiceNow ticket and use LLM, this is what is getting prompt to LLM.54:07So the quality of this prompt is gonna decide the quality of your, of your output or of your recommendation to you along with your ServiceNow ticket. So if you remember from our last session, Alan showed you how you can add documents to your mystery to improve it. It's, it's not.54:16So this is where if it's hosted on prem and you want to give it more context, maybe upload some troubleshooting documents from Cisco or if you have anything custom made, uploaded so that, you know, it has more context and it can give you more meaningful information from this prompt.54:36I'm gonna show you how it's look in just a few seconds, but before that let's go through all the tabs on the left side. So then we have ServiceNow integration, which is where you can add your configuration, then your email here.54:54Then you could test your webhook application, but we're gonna do it live here so we don't need to test the webhook, and then the logs that we have here. So you can see that this was something that at on 25 of March at a.55:11Around 02:00 P.M. it's it detected something and it created a ticket and then if you click on this I like icon, it will show you the LLM response that this error indicates that there's an IP address conflict in your network, this device with this, this, this, sending an ARP message, and then it's.55:31Giving you troubleshooting steps and it all depends on what kind of prompt you give. So if you just want troubleshooting step, if you just want to know more about what the problem is or if you wanna make how you want to present your, your text maybe in a table format or any other format.55:51So this is all depends on prompt prompt on your prompt, the response, and then it's gonna create your ServiceNow ticket and so much more. So now this was the last ticket that we created at 02:11 p.m.. So now we're gonna go to our Cisco modeling lab. I'm gonna ping something.56:10From our our data center client to to our default gateway which is 10011 and then it's gonna detect something on detect a syslog message on Nexus switch, let's see. So pinging it.56:30And then if we go to Nexus switches, this is what we see here. And now if I go back to my webhook application and look at the logs, I'm gonna refresh it. Remember that this is 02:11. And then it says something is processing because it has detected the this due.56:46Duplicate messages, so it's processing it's it's taking input from your LLM before generating your ticket. So if you look here, there is NO ticket generated yet.57:06And then if I refresh again, the status is completed and the ticket is generated. So this is our new one. If you click on actions, you will receive very similar actions because the prompt is about the troubleshooting steps.57:14So this, this use case is very useful in reducing the time to process your tickets, alerts, and then gives you recommendation about next steps. All depends on your near prompts. So the best part is you can use it on prem infrastru with on prem infrastructure. If you are worried about your security, your data leaving.57:33Your premises or or even the latency, but you can also use it with public cloud, hybrid cloud, anything. So with that, I'm gonna pass it to Riley to show you all the the game that we're gonna play and it is based on our hybrid model.57:53So over to you Riley. Awesome, thanks so much Priyanka and Joe. I'm gonna welcome everybody. We're gonna end the session today with our final demo and we're gonna have a little bit of fun here.58:11We're going to be playing a Lego game actually that is going to be hosted on Cisco AI bots. So let me go ahead and share the screen here and we'll get started.58:27Awesome.58:51So as you can see here, we are going to be using a Cisco AI pod to get ready to unleash your creativity in this fast paced. You'll get to build what is shown in the screen. You have 100 s to recreate it using only your imagination, quick hands, and clever design.58:53There are NO instructions to add more thrill. Our AI models deployed on our Cisco AI pod will give live commentary and score your creation based on design, creativity, and build quality. Let the building begin. Awesome, so it looks like I'll be playing to get some here and we are building a car so.59:10Well, let's go ahead and get started and good luck, guys.59:30Have some fund Welcome to the game where Riley, Mark, Priyanka will compete for 100 s to create the ultimate masterpiece.59:33Alright, so we'll be using some real life Mark is really bringing the yellow stuff together here, but I'm not sure that's gonna help him in the creative use of color category. Oh my Priyanka is going for a minimalistic approach, but will the judges find the lack of visible detail a little.59:51Lacking. Riley's really put some metal into his build, but I'm not sure the judges will be able to look past the lack of wheel wells. Priyanka is making a bold choice. I don't think i've ever seen a lavender car before. Mark is really thinking outside the box with this build.01:00:06Is that a koala? And is he driving? Riley is attempting a festive theme with the ornament, but I'm not sure he's gonna be able to convince the judges that it's a car. Mark's build is certainly yellow.01:00:26It's hard to miss that measuring tape. Oh my Riley appears to be attempting to pass off an ordinament as a car. What a Jolly idea? Oh mike. Priyanka is certainly going for a streamlined look.01:00:42But will the judges appreciate the lack of visible wheel wells? Oh Mike. Mark is really thinking outside the box with this build. Is that a coala? And is he driving?01:00:59Time is up. Carefully take your hands off the lego board and step away.01:01:24All right, we're all finished, let's see what the the judges have to say and the winner of our toy brick Car Building contest is. Priyanka! Congratulations on your fantastic creation.01:01:32Congratulations Priyanka, I like your car. Let's see if we can look at the scores here. Riley's core is six. Riley's creation while unconventional, shows some creative use of available objects.01:01:51The execution is basic, but the form of a car is somewhat present. Priyanka's score is seven. Priyanka put together a car build that is instantly recognizable. The execution is simple but effective, and the choice of objects is somewhat creative. Mark score is one. Mark.01:02:08Unfortunately, did not submit a photo, so therefore I have to give him a score of one. Oh bummer mark doesn't look like your image got submitted. But anyway, thanks for playing. That was a fun time and a good demonstration of.01:02:25You know, AI on the Cisco AI pods. So yeah, what we've just experienced was the live live AI inference in real time runs on a hybrid cloud setup. The Cisco A AI pods handle the fast localized inference for voice and image generation, and this is leveraging Gemini flash two point oh.01:02:41Model generating commentary and some orchestration.01:03:01Obviously this was just a game for fun, but the same methodology is actually gonna be, you know, applicable to multiple use cases. This is the same technology that we can use to, you know, watch our data center and then correlate physical activity with network events. So, this can be used for real time visual inspection and quality control and manufacturing.01:03:05It's a pretty great use case or self documenting video streams that automatically flag safety hazards. So fun element aside there's actually some real applicability here. And that about it does it for me. Thanks again guys for joining in that game and I'll pass it back to you Joe and you can wrap things up for us.01:03:25Great game guys. There's a lot of interaction you had there got the camera phones going and building cars. Wonderful.01:03:45Now for our audience, we'll take a look here. Raleigh did mention what the Lego game was using there behind the scenes. And for conclusion, we're gonna look at some key takeaways. Three of those 1st we've got.01:04:05That we want to select and prioritize our use cases. Now we've shown you some examples across different industries and you need to identify now what drives value in your specific context. You need to work with your team, your business leadership in order to define the top one.01:04:24One or two priorities and opportunities. Now look at what waste your team's time. It's also AI that delivers return on investment. The CX team and your partners here at Cisco can also be engaged to help you find and determine.01:04:44There's different use cases, specific to your environment. 2nd we want to ensure that your AI infrastructure is truly ready. When you look at do you have the networking, the compute, the storage, and the software stack that.01:05:02It's properly integrated and capable of supporting AI workloads. And that's where Cisco AI pods provides a validated, simplified path for you. Now MIT found it as three times more predictive of AI success when your infrastructure is AI ready.01:05:22AI projects that you're gonna run are gonna succeed or fail based on whether the infrastructure that supports them is ready. And thirdly, we want you to integrate and iterate. So as we said before, AI in a silo delivers nothing.01:05:43AI is connected to things like your ServiceNow, to your monitoring environment, to your different workloads. Those deliver value. You want to start small, you want to prove value, then you want to expand.01:06:03And what you saw today ran just on 2L40S GPUs, Nothing massive. It's practical, it's deployable and achievable. That's why we brought it to you today. If you're leaving here thinking about which use cases are most relevant for your particular organization, then we've even.01:06:19Achieved our goal.01:06:38We wanna spark those ideas, that's why we're here and to show you what is possible. With that, i'll turn it back over to our host Mark, take it away Mark. Thanks Joe and thanks Riley and Priyanka amazing session today and and some fun demos.01:06:40Route it. So for our next steps, we have our next session, which is going to be a Gentic AI from automation to autonomous networks. You can use the QR code to register for next week's event. I've also posted a, a link in the chat where you can register there as well.01:06:59We also have the Slido survey available. If you could take a second just to, to let us know how we're doing, we would really appreciate your feedback. And we also have the Slido Q and A available. So if you have any.01:07:20Questions, feel free to post your questions there and we'll make sure to get that answered.01:07:40If you have an idea for a topic that you would like for us to cover in a future expert insight series, you can also leave any ideas within the survey itself. So we're always open to hearing what our customers want to learn more about. So if you have an idea, feel free to to leave that within the survey.01:07:53Yeah, take a pause here to see if there's any questions. I'm not seeing any questions in the queue at the moment alright and with that, thank you.01:08:18To everyone for joining. We hope to see you on our next session. Have a beautiful day. This is our expert insight series on real life AI use cases for Cisco infrastructure. Thanks everyone. Have a good day.01:08:38