SREDAY

OpenTelemetry makes observability simple—add an agent and instantly see every service interaction. But this ease creates a hidden trap: teams instrument everything without considering the cost impact, leading to observability bills that consume 10-30% of infrastructure budgets. This talk exposes the real cost crisis behind modern telemetry data. We’ll examine actual cases where companies faced million-dollar annual monitoring bills, break down why high-cardinality metrics create exponential cost growth, and show proven techniques to cut expenses by 60-90% while maintaining full visibility. Through live demonstrations with real production scenarios, you’ll see how simple configuration mistakes create massive data volumes, then learn practical solutions: smart sampling that captures 100% of errors while reducing normal traffic by 95%, edge filtering that stops expensive data before it hits backends, and retention strategies that slash storage costs 60-80%. This isn’t about avoiding OpenTelemetry—it’s about using it smartly. You’ll gain actionable frameworks for cost-controlled implementation, team accountability models, and monitoring tools to prevent bill shock. Leave with battle-tested strategies from companies handling massive scale while controlling costs, plus a practical toolkit for cost-effective observability that finance teams will approve.

Mukta has survived 22 years of production outages, 3 startup exits and somehow still thinks observability is fun. She currently helps Last9 customers implement their Observability needs while keeping their sanity intact.

• OpenCost is an open-source tool designed to measure the granular cost of cloud resources.
• It is best suited for cloud-native environments, but works across several cloud platforms.
• While everyone knows the basics of cost optimization, few apply them proactively—they’re usually implemented in retrospect.
• The primary reason for this reactive approach:
• Lack of situational awareness
• Insufficient insight into granular resource costs
• Engineers can estimate total costs for compute instances and infrastructure stacks, but struggle with:
• What does a build cost?
• What does a deploy cost?
• With the rise of AI workloads, costs are escalating.
• More responsibility falls on SRE teams to estimate costs accurately and provide better cost visibility.

Ram Iyengar is an engineer by practice and an educator at heart. He was (cf) pushed into technology evangelism along his journey as a developer and hasn’t looked back since! He enjoys helping engineering teams around the world discover new and creative ways to work. He is a proponent of product development and engineering teams that put the community first.

In an era of increasingly complex, distributed systems, Site Reliability Engineering (SRE) teams are perpetually navigating a deluge of operational data, often battling manual toil, fragmented tools, and the relentless pressure of incident response. The traditional approach, while critical, often leaves SREs in a reactive stance, hindering proactive innovation and true operational excellence.

This session will explore the transformative emergence of SRE AI Agents – intelligent, conversational systems designed to augment human SRE capabilities and fundamentally reshape how we build, monitor, and maintain reliable systems. We will delve into how these AI agents, powered by advancements in Large Language Models (LLMs) and sophisticated data correlation, are moving beyond simple automation to become true "co-pilots" for reliability engineers.

I am passionate about building reliable systems that just work—for everyone, every time. Currently leading global Site Reliability Engineering efforts at Ford Motor Company, I focus on ensuring our Global eCommerce Platform is not just operational but resilient.

Modern systems are too dynamic for dashboards alone and too business-critical to leave to black-box automation. This session reframes “observability 3.0” as a sociotechnical practice that fuses human-driven SRE with pragmatic AIOps. We’ll explore how rich telemetry (metrics, logs, traces, profiles, and user signals) plus topology and domain context evolves from passive visibility into an active decision system. On the human side, SREs define intent via SLOs, failure hypotheses, and reliability narratives and design for explainability, guardrails, and post-incident learning. On the AI side, we’ll discuss how correlation, anomaly detection, event deduplication, and LLM-powered triage can cut noise and accelerate root-cause discovery, while runbook automation and safe auto-remediation handle the boring, reversible fixes.

Expect concrete patterns: telemetry pipelines that feed feature stores, knowledge graphs that reflect real service topology, feedback loops that keep models honest, and governance that ensures AI augments not overrides SRE judgment. We’ll also cover common traps (data debt, alert fatigue, overfitting to incidents) and a rollout playbook that starts small and proves value quickly.

Seasoned, forward-looking professional with around two decades of experience in Site Reliability Engineering, Solution architecting, Performance engineering, Capacity planning, Chaos Engineering, product development, automation and setup SRE for large enterprise applications in various business domains. Consistently recognized as competent individual, skilled at coordinating with cross-functional teams in a fast-paced environment to steer timely completion of project with budgetary constraints.

This is a story from the trenches of running one of India’s largest ECS fleets—serving millions of requests entirely on infrastructure that can disappear with just two minutes’ notice. We began with the “easy path” of a third-party managed solution, but as we scaled, it quickly became our biggest bottleneck and a massive cost center.

This session is our journey of taking back control. We’ll share the hard-earned lessons from hitting undocumented AWS limits, battling against opaque “black box” algorithms, and enduring production outages that forced us to innovate.

You’ll learn how we:
- Built custom controllers and predictive scaling
- Slashed instance boot times by 75% through EBS optimization
- Developed a “25-second miracle” shutdown process

All of this allowed us not just to survive—but to thrive—in the chaos of a 100% Spot environment, achieving 99.99% uptime while cutting compute costs by 60%.

This is how we transformed ECS from a simple orchestrator into a battle-hardened, intelligent platform.

Gaurav Chauhan is a Software Development Engineer at Zomato, with prior experience at Oracle, CRED, and Capital 2B. An alumnus of IIT Delhi, he has worked across backend systems, data science, and cloud infrastructure, and has held leadership roles driving large-scale initiatives.

Sudip Chakraborty is a Software Development Engineer at Zomato, working on the platform team with a focus on reliability, benchmarking, and large-scale microservice optimizations. Over the past three years at Zomato, he has contributed to ML Ops, backend systems, and platform engineering, leveraging technologies like Golang, Kubernetes, Terraform, and Kafka.

• Debugging Kubernetes workloads across ephemeral pods, nodes, and services is not straightforward, especially in real time.
• Traditional observability tools provide high-level metrics, but lack the depth and immediacy needed to find root causes of elusive, low-level issues.
• This is where eBPF and Inspektor Gadget excel.
• Inspektor Gadget leverages eBPF to dynamically instrument the Linux kernel, providing:
• Real-time insights into what’s happening inside a Kubernetes cluster
• No need to install agents
• The talk will cover how to use different gadgets from the Inspektor Gadget toolkit to observe:
• TCP connections made by a pod
• DNS requests
• POSIX signals
• Syscalls
• Files opened by a pod
• traceloop — a flight recorder for syscalls
• And more
• Structure of the talk:
• Introduction to eBPF
• Overview of the Inspektor Gadget suite
• Live demos of different gadgets in action
• Key takeaways for the audience:
• How to simplify debugging difficult Kubernetes issues with Inspektor Gadget
• How to capture output in a user-friendly format for later parsing

I am a senior SRE at Sematext, where I am responsible for entire AWS/Kubernetes infrastructure that is used by hunderds of customers worldwide. We self host almost everything that we run, including Kubernetes. I love to solve challenging infrastructure problems with a focus on automation and reliability. I have spoken in several conferences, like DevopsDays, Cfgmgmtcamp before, including local meetups.

Conversational AI has captured widespread attention, but the true potential of AI lies well beyond chatbots and dialogue systems. As enterprises look to build intelligent, adaptive systems, the ability to provide real-time, contextual understanding becomes essential. This is where Model Context Providers (MCPs) come in — systems designed to dynamically supply AI models with the contextual signals they need to make relevant, timely, and intelligent decisions.

In this talk, we’ll explore the evolving role of MCPs in modern AI architectures and how they can be used not just to improve inference quality, but also to intelligently manage and orchestrate Kubernetes environments. By integrating context-aware models with cloud-native infrastructure, organizations can unlock powerful capabilities — from self-tuning systems and adaptive resource allocation to AI-driven automation across the stack.

We’ll dive into: - What MCPs are and why they matter beyond chat interfaces - How to build MCP server - How MCP can be used for managing Kubernetes cluster - How to integrate MCP with tools vscode, Claude, etc.

Aditya is a Senior Software Engineer at Walmart. As a proud CNCF Kubestronaut, he holds multiple Kubernetes certifications that showcase his deep expertise in the ecosystem. Beyond his work, Aditya actively shares his insights through his YouTube channel, creating tutorials on cloud technologies and software engineering, and writes technical blog posts to help others navigate and master these domains.

This session is for SREs, DevOps engineers, and platform teams who want to strengthen Kubernetes security at the network layer. While cloud providers offer firewalls and service meshes, the last line of defense inside the cluster is Network Policies.

The talk balances concepts + live demos and provides a clear journey: starting from the “default open cluster,” then step-by-step applying Network Policies to enforce strict communication.

I will use Calico on GKE to illustrate examples, but the learnings apply to any Kubernetes distribution. The session ensures attendees leave with concrete policies they can apply to their workloads.

Sanket is a working as a Sr Devops Engineer at Lloyds Banking Group and He is a Google Developer Expert in GCP. He has worked as a cloud engineer at Searce (GCP Premium Partner) He is Google Cloud Champion Innovator in the category of Modern Architecture with Certified Kubernetes Administration (CKA) , 3X Certified in GCP and 1X in Azure . He has been helping small and mid sized startups to adopt and implement best practices in cloud and DevOps culture to fasten their software delivery process. He loves to integrate various GCP service like Vertex AI, Agents Development Kit, MCP servers,networking, compute, storage , containers, GKE etc, and likes to Implement various use cases. He loves to explore and deep dive into GCP services and also help community by creating content and writing medium blogs.

Monitoring and observability at enterprise scale is challenging — especially when dealing with 500 VMs and 3 Kubernetes clusters totaling 1,500 nodes. In this session, I’ll share how we implemented full-stack observability using Coroot agents, collecting metrics, logs, traces, and CPU/memory profiles out-of-the-box with minimal operational overhead.

I'm passionate about bridging technology and business outcomes. I have over 20 years experience in Performance Engineering and Site Reliability Engineering. As a founder of CloudElu Labs, I help engineering and product teams embed SRE, FinOps Chaos Engineering and Observability practices directly into their DevOps workflows — starting from development.

I also founded From Dev to Ops - a Chennai based Devops Community, which brings together practitioners of Devops, SRE and Platform Engineering to Learn and Share the expertise.

With a strong belief that visibility drives innovation, I specialize in implementing cost-aware infrastructure, OpenTelemetry-based observability platforms, and developer-friendly tooling across AWS, Azure, and GCP.

I work with forward-thinking leaders to eliminate blind spots in cloud spend, improve reliability through modern SRE practices, and enable strategic, forecast-driven cloud decisions — without adding friction to engineering.

In modern applications, observability is essential. OpenTelemetry (OTel) has emerged as the standard for telemetry data, providing a unified way to collect, process, and export logs, metrics, and traces. Beyond data collection, organizations need effective ways to store, analyze, and visualize this telemetry to drive actionable insights.

This session is for developers looking to deepen their understanding of OpenTelemetry's architecture and its three key pillars: logs, metrics, and traces. We will walk through how OTel fits into an observability stack, the benefits it brings, and common integration patterns. A small demonstration will showcase how telemetry data can be captured and visualized in practice.

Audience Takeaway: A clear understanding of OpenTelemetry’s core components, how it supports observability across modern applications, and hands-on insights through a live demo.

Advisory Software Engineer at IBM Labs with a passion for Observability, APM, IoT, and Automotive solutions. A 3x Patent Holder and innovator in telemetry and distributed systems, specializing in Microservices, Kubernetes, and Cloud technologies. Expert in Java, Kafka, MongoDB, ElasticSearch, and Cassandra. Committed to building scalable, high-performance solutions that drive real-time insights and enhance system reliability.

```markdown Why should SREs care about systems thinking applied in aviation safety engineering?

Modern distributed systems face the same challenges as complex safety-critical systems: emergent failures, cascading outages, and the gap between system design and runtime behavior. While traditional monitoring focuses on known failure modes, aviation safety engineering provides systematic approaches to identify unknown risks.

In this talk, you will learn how we applied MIT's System-Theoretic Process Analysis (STPA) — a methodology from aviation safety — to analyze reliability risks in a large-scale eCommerce platform processing millions of transactions daily.

What you will learn:
- How to map STPA concepts (hazards, constraints, control actions) to distributed systems components
- A systematic framework for identifying cascading failure scenarios before they occur
- Practical techniques for analyzing interactions between auto-scaling, load balancing, and circuit breakers
- How cascading failures were reduced using insights from this analysis
- Actionable methods you can apply to your own systems

Real examples covered:
- Circuit breaker coordination failures that created retry storms
- Auto-scaling feedback loops that amplified rather than dampened failures
- Security policy interactions that blocked legitimate traffic during incidents
- Configuration drift detection that prevented silent reliability degradation

This is not theoretical — we'll show concrete code examples, architecture diagrams, and actual incident data. You'll leave with a practical toolkit for systematic reliability analysis that goes beyond traditional SRE approaches.

Whether you are dealing with microservices, serverless architectures, or hybrid cloud deployments, this methodology will help you build and maintain more resilient systems by thinking systematically about failure modes and control structures.

Perfect for: SREs, Platform Engineers, and Engineering Managers who want to move beyond reactive incident response to proactive reliability engineering.
```

Mahesh Venkataraman leads innovation in the area of application of artificial intelligence, data mining and machine learning in software engineering. He has led successful implementation of natural language processing driven test automation, usage and failure modeling using log analytics, empirical analysis of technical debt and application of knowledge graphs in discovering patterns and relationships for optimizing test suites and improve decision making for system integration projects. His passion is bridging the gap between theory and practice, between academia and industry and creative thinking in software. He is a regular keynote speaker in many conferences. He is currently working on addressing uncertainty in fault prognosis and diagnosis

Koushik Vijayaraghavan is a Senior Managing Director at Accenture, where he has spent over 20 years driving product innovation, engineering, and digital transformation for global clients. He began his career at Cognizant and has completed Harvard Business School’s Disruptive Strategy program, strengthening his expertise in guiding organizations through change.

Why can't you just use your old monitoring tools for your new AI systems? This session answers that question by introducing AI Observability. You'll learn how to use AIOps to move beyond guesswork and effectively manage your AI agents and infrastructure

Meiyappan Kannappa is a Technical Director at Ford Motor Company with around 20 years of experience in designing and developing architectural frameworks and cloud applications. He specializes in modernizing applications to cloud-native technologies and has extensive experience building B2C and B2B systems within the e-Commerce, Connected Vehicles, and mobility sectors. As a technical enthusiast, he actively writes about innovative software architecture for digital transformation.

I’ve engineered an AI-powered Continuous Performance Testing pipeline: JMeter unleashed in GitLab CI/CD, Python-driven baseline magic, and Google Gemini LLM delivering smart insights. Real-time, action-ready alerts hit Slack—be the first to squash regressions! This talk reveals every step

Ganesh Padmanaban is a Principal Site Reliability Engineer at Guhatek with over 15 years of experience helping organizations improve the reliability and performance of their systems. He has led teams in streamlining how software gets tested and released, making processes faster and more efficient. Ganesh is known for introducing new approaches that use automation and advanced AI tools, resulting in quicker releases and more dependable systems. As a certified expert and active member of the tech community, Ganesh is passionate about making reliability easier for everyone and shaping the future of how companies build and maintain their technology

What We’ll Cover

• Massive Scale Orchestration
• Executing 1M+ CI/CD jobs every month across 600+ repositories in multiple GitHub organizations.
• Each workflow run triggers comprehensive validations, combining open‐source tools and in‐house linters.

• Build-intensive repositories consuming 60–70 CPU cores per run, showcasing CI/CD performance at extraordinary scale and speed.

• Solving Critical Edge Cases

• Managing spot instance terminations seamlessly.
• Mitigating GitHub API throttling through dedicated Apps and bots.

• Fortifying orchestration against cascading failures caused by unpredictable GitHub service outages.

• Cost Optimization

• Running CI/CD at up to 10x lower cost compared to GitHub-hosted runners.
• Leveraging a mix of spot and on-demand machines with smart bin-packing.

• Implementing intelligent workflow filters to eliminate unnecessary runs and maximize efficiency.

• Performance Enhancements

• Leveraging caching of packages and builds to speed up performance and reduce redundant work across workflows.

• Operational Standards

• Rolling out changes across multiple environments with robust standards, ensuring stability and reliability.

Key Takeaways

For Engineers and Operators: - Scalable Infrastructure
Design ephemeral ECS-based self-hosted runners that handle massive CI/CD workloads effortlessly.

• Cost & Performance Optimization
  Unlock strategies to reduce costs and maximize efficiency with self-hosted runners.

• Accelerated Job Runtimes
  Improve provisioning speed and leverage caching for faster workflows.

• Automation & Standardization
  Enforce linters and automate standards across repositories.

• End-to-End Deployment Cycle
  Build deployment workflows with safeguards, visibility, and approval gates.

• Advanced Reliability Features
  Implement automation to enhance reliability during deployments while reducing human intervention.

GitHub Actions as a Unified Platform The talk emphasizes GitHub Actions as a single platform for CI/CD, streamlining workflows, improving developer experience, and eliminating the need for multiple tools.

By sharing Zomato’s journey, we aim to inspire practitioners to rethink and enhance their own CI/CD setups—making them more resilient, efficient, and developer-focused.

Akshat Goel is a Software Engineer II at Zomato with over three years of experience in building scalable systems. He holds a Bachelor's degree in Computer Science from IIT Ropar (2018–2022) and has a strong background in backend development and distributed systems.

Nishant Sarraff is an SDE-2 at Zomato with over three years of experience in developing scalable software solutions. He earned his B.Tech in Computer Science from IIT Jodhpur (2018–2022) and has expertise in backend systems and performance optimization.

Kubernetes promises portability and scalability, but in reality, most production outages happen due to avoidable mistakes. Security gaps, misconfigured health checks, poor scaling strategies—all can derail even experienced teams.

In this session, we’ll uncover:

Security Disasters → The risk of running containers as root, overlooked image vulnerabilities, and RBAC pitfalls.

Configuration Catastrophes → Why “works on my machine” never works, and how resource mismanagement wrecks clusters.

Observability Blind Spots → Missing runtime security monitoring, misleading CPU/memory metrics, and logging anti-patterns.

Scaling Traps → HPA-induced thrashing, node scheduling inefficiencies, and bottlenecks hidden until too late.

But it’s not just about problems—we’ll explore solutions:

Actionable hardening checklists

Tools for continuous monitoring & runtime security

Automation strategies to prevent config drift

Proven observability practices for anomaly detection

Whether you’re new to Kubernetes or running enterprise-scale clusters, you’ll leave this session with practical, battle-tested strategies to keep your systems safe, stable, and observable.

Kaustubha Shravan is a Cloud Architect who designs and operates resilient, measurable, and cost-efficient platforms across Azure, AWS, and GCP. She blends reliability engineering with data-driven practices—SLOs, error budgets, and ML-assisted incident response—to make outages rare and recovery fast. With 46+ cloud certifications, she has led initiatives such as Benchmarking-as-a-Service and production-grade ML inference pipelines that improved performance while cutting spend. Kaustubha is a Women Techmakers Ambassador and frequent community mentor; her work has been showcased at NeurIPS workshops. She speaks about pragmatic reliability patterns, observability that drives action, and culture—how to turn postmortems into durable engineering improvements. When she’s not shipping guardrails, she’s helping teams adopt sustainable, privacy-aware AI practices and sharing playbooks that teams can put to work immediately.

Imagine an AI Agent that detects production incidents from logs, finds a resolution runbook and resolves it autonomously. In this session, we’ll walk through the Strands Agents SDK and show how to create such an agent. We will use the RAG pattern to create a repository of Runbook knowledge. Then take it one step further by enabling the agent to call the tools so that the agent can resolve incidents autonomously. Increasing your peace of mind and accuracy, and reducing the MTTR at the same time. Walk away with actionable insights and a demo.

Manu is a seasoned software professional with over 20 years of experience in the tech industry, spanning enterprise applications, cloud solutions, and data-driven platforms. As an AWS Community Builder for 3 years and a holder of multiple AWS certifications, he brings deep expertise in designing scalable, cloud-native architectures. A passionate advocate for AI, data, and automation, Manu focuses on practical applications of emerging technologies. When not building or mentoring in the cloud space, he can be found in his kitchen whipping up a delicious cake.

Let's be honest: "blameless culture" is easy to say, but hard to do. We've all been in those post-mortems where the air is thick with unsaid accusations and the search for a root cause feels more like a search for a scapegoat. What happens when good intentions clash with the messy reality of a production outage?

This talk is our story of that journey. We'll share the gritty details of how we built a culture of psychological safety from the ground up. This isn't about theory; it's about the practical systems we put in place, from our rotating "First Incident Responder" role that turns on-call into a powerful learning opportunity, to our weekly Incident Response Reviews, a dedicated forum where engineers openly deconstruct failures without fear.

I love collaborating with cross-functional teams to solve complex challenges and leverage disruptive technologies to drive innovation.

As the leader of the DevSecFinOps team at Nuclei, I have honed my ability to empower my team as well as expedite value delivery of our products by intuitively architecting robust strategies and tactically executing them via cutting-edge tech stack that is highly available across multiple zones and is disaster resilient across multiple regions based on active/passive with pilot light disaster recovery strategy that has an RTO of <10m and an RPO of <10ms.

My infinite mindset and pragmatic approach helps me to swiftly identify and prioritize solutions that addresses current business need as well as aligns with the long-term vision. I am committed to consistently expanding my horizons and delivering high-quality results that always exceed expectations.

Ever wonder how top teams keep complex systems running smoothly? This keynote will explore the current landscape of observability, moving beyond traditional pillars to embrace advanced techniques and holistic insights. We'll discuss how SRE teams are leveraging emerging technologies to navigate increasingly complex systems, transforming reactive firefighting into proactive engineering.

Hemant Kamatgi is a technology leader with over two decades of experience driving large-scale engineering and digital transformation initiatives across retail, eCommerce, and automotive industries. He currently serves as Vice President of Customer Platform Engineering at Ford Motor Company, leading global platform strategies from Bengaluru. Previously, he held senior leadership roles at JCPenney, Reliance Retail, Target Corporation, and Cognizant, building and scaling customer-focused platforms and engineering organizations.