Hi everyone. I'm Manisha Otti. I'm so glad you're joining me virtually today I work on designing modern cloud-based systems that blend artificial intelligence with strict regulatory needs, especially in healthcare, banking, and insurance. Today I'll share how we can build, aI systems that are fast, scalable, and still safe and compliant. We will explore how Kubernetes helps us automate compliance, cut manual work, and build a trust in ai. Let's start with the real world problem we are trying to solve. In almost every organization I have worked with, AI is everywhere. Predicting risks, automating tasks, personalizing services. But in industries where lives or money are at stake, AI has to be responsible. Let me give you a few examples. In healthcare, we can't just predict. Patient outcomes. We must show why the model made that prediction. In banking, fraud detection has to be fast, but also explainable to regulators. In insurance, every automated claim decision must have a trial that says who change it, what, when, and why. So the paradox is simple. AI wants speed. But regulators want control. So how do we innovate without breaking the rules? That's where Kubernetes comes in. So think of Kubernetes, like Canada, a traffic controller for your applications. It tells each plane, your AI model, where to go. When to take off and how to land safely. So this brings four big advantages, self-healing. If a model crashes Kubernetes restarts it automatically. Automation, you define the desired state and the system keeps it that way and security. Workloads are isolated. So sensitive data, stay safe. Cloud independence run everywhere on premises or in the cloud. For example, in one of our projects, a healthcare data matching service used to take engineers hovers to redeploy if something failed. After moving to Kubernetes, failures fix it themselves within minutes. Zero human action. Let's ground this in three sectors where compliance truly matters. Healthcare. Imagine a global clinical trial platform analyzing the patient data. Daily rules require full privacy encryption and also the audit trials. Also for banking, a fraud detection model flags suspicious transactions in milliseconds, but it must also store why each flag was raised Also for insurance and underwriting. Engine predicts risk, but must explain its reasoning to both regulators and also to the customers. In all this, Kubernetes helps maintain isolation, separating the sensitive data and applying security rules. And also on top of it, keeping the deployment history for every version. Here is a simplified, look at the architecture, like how we use each AI model is packaged as a container, like a neat little box holding everything it needs. Kubernetes, deploy, deploys these boxes managing version upgrades automatically and state full sets like which. Makes sure models that store data can roll back safely if something breaks. Then comes key tops where every configuration and model change is stored in kit. That means when an auditor asks, who approved this change, you can show them instantly. We once had a regulator request the exact deployment history for a patient data matching model. Because of GitHubs, it took us seconds to show, not days. With Kubernetes complaints is in intent afterthought, but it's part of the design. The main is a. Security rules like force containers to run in safe read-only mode. Network policies, block any communication, not explicitly allowed. Admission controllers reject anything that violates policy before it ever runs. And also the persistent logs automatically stores the Uneditable audit trials. So instead of humans manually verifying every release, the system does it. For example, when one developer tried to deploy a model that wasn't code reviewed, the controller blocked it automatically. So that saved us from a compliance breach. Tops turns the compliance into automation. Every change starts as a pull request and automated checks validate security fairness and data rules. Then once approved, Kubernetes applies it safely and traceable. If something goes wrong, we can roll back instantly in one bank's deployment. Updates that use two take five hours now. It's taking just 12 minutes with zero manual steps. Every log, every version, every approval is automatically captured. Visibility builds trust, so we use service mesh technology to watch how each service talks to one another, just like a traffic camera system. This lets us do gradual rollouts and instant rollback. If something misbehaves for explainability, we integrate simple model explanation tools that visualize what influencer each prediction. So when a doc, when a doctor asks, why was this patient flagged high risk, the system can actually show the reasons. Monitoring dashboards powered by Grafana, and similar tools give live insights into a performance and cost. Everything is transparent. AI systems can burn money fast, especially with the GPUs. Kubernetes helps by scaling GPU, no notes when needed. Assigning GPUs to critical jobs. First, using cheaper preempt servers for bragger background tasks. In one insurance project, we cut compute costs by 40 percentage simply by letting Kubernetes decide where to turn GPUs on or off. We built a global patient record matching system before Kubernetes matching was slow and error prone as well. So after migration it handled 10 million requests daily and inference speed improved by 73% Compliance. It's fully automatic. The system enforced healthcare privacy policies at the deployment and stored immutable logs for audits. It's a great example of performance and compliance coexisting. A multinational bank used manual deployments that took covers and were error prone. After moving to Kubernetes pipelines, releases drop put from hovers to just minutes. Fraud detection became four times faster, and traceability satisfied the payment industry regulators automatically. So speed and safety in harmony. For a larger insurer, running models across regions, data isolation was the biggest concern. We used Kubernetes namespace to separate clients auto-scaling, to handle claim traffic, and built in explanations for every AI decision. The result is cost savings. Transparency and happier regulators. So these are the patents we follow repeatedly. Complaints as code like policies or return like code submissions, enforce them, immutable audit trails, logs can be altered. Safe rollouts. New models start with 5% of traffic. Then scale layered. Security training and serving are fully separated. Following This ensures consistent success no matter the industry. After years of deploying these systems. Here is what stands out. Automate compliance early. Build visibility from day one. Scale smart. Avoid over provisioning. GIS is a cultural shift. It blends transparency and also the trust. These lessons came from real teams, not just theories. What's next? We are moving toward policy driven runtime compliance where systems self-correct instantly. If a rule breaks, we are exploring serverless GPUs, pay per prediction instead of perha, and we are testing federated learning training models in multiple locations without moving the private data. The future is compliance that's intelligent, automatic, and invisible. To wrap up, Kubernetes gives us scale, compliance and efficiency. We have seen 90% faster deployments, 10 million daily requests, and four times faster detection. All in regulated environments, compliance doesn't slow innovation. It strengthens it. When it's built in, it becomes your greatest advantage. Thank you all for spending your time. I hope this session helped you see that compliance and innovation can truly work together.