Hello everyone. Thank you for joining me today. My name is Jmir Sink. Today I'll be sharing how predictive manufacturing analytics can be scaled using Kubernetes powered CRM systems. The big idea is this, when companies move from reactive service models to predictive proactive ones, they can unlock massive value, including the. The potential for dramatic revenue growth. Now let's dive right into it. I'm gonna start sharing my screen. Manufacturers face a big challenge after market services. Things like maintenance and spare parts represent a huge revenue opportunity, but most companies don't capture it effectively. Why? Because they're reactive. Equipment breaks, then they respond. Even with digital initiatives, many can't scale predictive service models, but if organizations shift towards predictive analytics driven operations, they can move from firefighting to creating real time enterprise scale value. Traditional CRMs weren't built for this. They were designed for tracking customer interactions, not billions of iot signals. They're monolithic, so scaling is hard. They struggle with velocity. The speed of realtime streams and volume of customer data explodes, and it all leads to a lot of problems. So if manufacturers want predictive insights, they can't just rely on old CRM architectures. They need systems that scale natively. This is where Kubernetes really changes things. By moving to Kubernetes native architectures, we can break analytics into containerized, microservices, small modular pieces that are easy to reuse and scale. When demand spikes, Kubernetes automatically scales resources up when demand falls. It dials them back down. That means performance when you need it. Cost saving when you don't. So what does a realtime pipeline actually look like? It starts with iot ingestion, sensor streaming, nonstop capturing, captured by containerized collectors. Then comes. Service history, all the maintenance logs and repair records analyzed in distributed fashion to surface patterns. On top of that machine learning models, inside Kubernetes pods model, customer behavior and usage trends. Finally, predictive insights are delivered in real time through auto-scaling services. So you're taking raw sensor noise, combining it with history and behavior, and turning it into actionable insight. If we go a little deeper into the architecture, we've got containerized pipelines with helm managed machine learning deployments, auto-scaling services, keep things responsive. While even driven architectures react instantly as data comes in on Kubernetes itself, jobs handle batch analytics, CR jobs schedule recurring tasks, custom operators streamline machine line workflows and state full sets manage persistence. Each piece has a role, and together they form a flexible backbone for predictive analytics. When companies design systems like this, the impact is clear. Customers stick around longer because service is proactive. Downtime drops sharply thanks to predictive maintenance, and most importantly, service revenue grows because resources are allocated at the right time in the right way. There are a few orchestration strategies worth highlighting. Use persistent volumes for stateful data, so context isn't lost. Distribute model training across multiple pods so it scales efficiently and use event driven triggers inside Kubernetes. So predictions handle happen the instant data arise. Resource allocation is where performance really lives or dies. CPU allocation should be dynamic. Workloads get compute power when they need it. Memory should be managed with limits and affinity rules, so no single pod hogs, the cluster and for storage. Persistent volumes handle model artifacts you need to keep. While a femoral storage takes care of temporary workloads, A typical orchestration flow looks like this. Data comes in from manufacturing systems, goes through EDL for cleaning, then analytics for feature engineering and inference. Finally, predictions get pushed out through APIs, into CRMs and dashboards. Simple flow data comes in, gets cleaned and enriched, and delivered. Of course, monitoring is critical. On the application side, you should track accuracy, drift latency, and data quality. On the infrastructure side, it's about monitoring. Monitoring PO resource use, scaling events and storage performance. Without good monitoring, you won't catch issues before they spiral. There are pitfalls to watch for if you don't set proper resources. Resource requests and limits workloads will compete and slow each other. Each other down if you don't design data access. IO bottlenecks creep in fast, and if state isn't managed properly, models and intermediate data can disappear during scaling. These are avoidable, but only if planned for upfront. Finally, resilience production systems should be able to handle failure without breaking. That means multi-zone deployments, circuit breakers for fall tolerance and horizontal scaling policies to balance workloads automatically. And of course, persistent volumes and backups. Keep data safe. Resilience is what turns and experiment into something the business can rely on. So to wrap up scaling predictive manufacturing analytics isn't just about analytics, it's about architecture. Kubernetes native arch native systems give us the scalability, efficiency, and resilience needed to move from reactive service to predictive operations, and that's where the real business impact comes from. Thank you so much for joining me today. I'd be glad to continue this discussion and answer your questions after this session. Thank you.