Hello everyone myself, Ika Ram. Welcome everyone. So today I will be walking you through how AI and machine learning are changing instant management in the cloud. For over the years we have been stuck in the reactor mode waiting for outages to happen and then scrambling to fix them. What I'll show you today is how distributed AI architecture, observability, and automation pipeline can change that. Focus in both strategic why and also why this matters and the technical and also how to build into your stack. So next slide would be today's agenda here. How we will structure the session first. We will look at how instant management has evolved over, over the time. Then we'll break down the core AI and ML capabilities that driven modern systems like things like an detection. Root cause analysis and auto autonomous redemptions. Next next I'll walk you through distributor AI architectures where telemetry models and automation pipeline come together. So finally, we will wrap up with the real world case studies with key metrics and step by step roadmap so you can use for implementation. So next slide would be. The high stakes of modern incident management. The stakes today are higher than ever. In cloud environment downtown downtime can cascade through dozens of services and cost millions of dollar per hour. Incidents are still caused by human error and alert feature tube. So where engineer get so many false alarm that critical signal get lost. The complexity of distributed microservices only multiplies the problem. The truth is manual approach doesn't scale anymore. So we need intelligence system that can shift. Through nausea, identify through issues and guide engineers towards the right action as a faster, okay next slide would be the evolution react to pro two. So let's break down this evolution, like in past instant management was reactive. The Sure based. Monitoring would trigger alarm when metrics like CPR memory crossed a limit and human had to dig through, log to diagnose the issue. Over the time we moved to more responsive system like correlation engines automated runbooks and structured post model. But the real leap forward is predictive ance. So where a model forecast alarm alarms before they turn into outages. Coming to the LSTM networks and time series forecasting can predict failures, like casual interference can pinpoint the dependencies and refor reformance learning agents can handle redemption automatically. The shift is from reacting after the fact to preventing issues before they impact the users. So next would be, yeah. Next would be the, oh my bad. Embarrassing. A new era. So this new era of intelligent incident management changes the entire life cycle, like metrics, logs, and traces are streamed in real time. Into a unified observability platform, whereas in machine learning models continuously analyze that data form an patterns and correlations. Instead of waiting for a human to review dashboard, the system can detect the issue and map the root cause, and in many ca cases, apply the fixed automatically. What this means in practice is less downtime. Lower co operational cost and higher customer satisfaction. Next slide would be core AI or MI capabilities. So these are four key AI capabilities at play here. The first is predictive validating, whereas model trained on baseline so it can detect early warnings signals long before traditional threshold are breached. The second would be is automated Root Task Analysis. So it's a craft-based model. Look at service dependencies and isolated with the fault originates. Where coming to the third is an autonomous redemptions where this is reinforcement learning agents can restart the services, reroute the traffic or scale, cluster on their own. And finally. Intelligent escalation where incidents are routed automatically to the right engineer based on the expert context and availability together. So these capability reduce the noise and cut detection and response times and. Let human focus on the high value work. Coming to the next slide where it driven to the leading platform. So there are the major cloud cover already embedded this capability AWS SageMaker, which integrates with the CloudWatch so it can deploy and it detects model and trigger the Lambda functions for. Automated redemptions, whereas coming to the Azure ml. Ml, so it ties into the log analytics and log apps where it detect can direct directly trigger correct corrective actions. But coming to the other cloud-based vertex AI connects, which with the cloud operations and cloud functions for event driven responses. While it's ML ops tooling keeps model restrained on fresh telemetry. So each provider is making AI operational intelligence a first class service. Coming to the next slide, like distributed AI architecture here, how the architecture looks in practice Telemetry is a collector from across your stack using open telemetry. Friendly or native cloud agents, a unified observability layer like elastic or datadog which correlates metrics, logs, and traces the a, these AI process layer. Runs the ML model for anomaly anomaly detection. We are forecasting and root analysis root cause analysis, typically using a distributor framework like Spark, ml or TensorFlow serving. So in response to orchestration, then connects to the ISTM tools like ServiceNow or pay page duty to trigger automated runbooks. A feedback loop retrains models continuously and secure data fabric ensure governance and compliance. This slide talks about it and coming to the next slide, that training intelligence. So the effectiveness of these systems comes down to the data. You train them on telemetry streams, provide metrics like latency throughout, through output and saturation locks can be praised with NVP techniques to extract error codes and correlation id. And coming to the torical incident record like post-mortem cells as a why training sets for root cause models. But the most important factor is data quality. So model trained on noisy. Inconsistent data will produce false positive. So observability pipeline must be cleaned and normalized before training. So coming to the next slide where it talks about the real world impact. So when origination adopt intelligent incident manager management, the numbers shift dramatically. We consistently see 82 and 90% reductions in meantime to detect and resolve incident. False alerts drop significantly ly, which reduces engineer burnout and attrition. So auto redemption rates climb as reinforcements. Learning models get better at handling repeated issues. So let me give you one example on the financial services side. For example used the predictive models to detect database saturation like 20 minutes before failure, so it automatically scaled the cluster and avoid a multimillion dollar outage. So these are the kinds of outcome that makes a business case clear. Next slide would be the implementation roadmap. Whereas coming to this slide like adoption work best in phases like first phase is data readiness, like centralized telemetrics, normalized logs and train baseline anomaly detection model. In phase two, we add augmentation and automation where it is deployed. AI driven alert, correlation, and run box. So coming to the phase three we introduce a proactive intelligence where it is a graph driven based root cause analysis and reinforcement learning for the redemption. Finally the fourth phase which is focused on the optimization predictive capacity planning like it's maintaining and expanding across region and the teams. So each step builds a capability while limiting its risk. Whereas coming to the next slide, it talks about the critical success factor where technology is only after equation. Success depends on high quality data, pipeline, clean and label for training, it requires a cross-functional collaboration like. Machine learner learning engineers, SREs and domain expert working side by side where it rollout should be iterate, like start with one service, do value, then scale and evolve. AI should augment human, not replace them. So exp expandability is key, like engineer must trust AI recommendation and or they won't adapt it. The future. Next slide would be the future is intelligent relations. The future is moving towards adaptive self uhhe system. ML agents will forecast failures, autoscale infrastructure and rero traffic before user ever notice a problem. Engineer won't disappear. They'll shift from firefighting to higher level design and governance. The organization that invest now will defend the operational religions standards for the next decade. The question isn't if the DA change is coming, it's whether you can lead it or follow it. That's the roadmap for, inte into the instant management, so thank you.