While LLMs make AI agent creation seem simple, building reliable production solutions requires structured methodology. This talk shares key insights from my year-long project developing an AI-powered root cause analysis system for Kubernetes clusters. I'll present three strategies that transformed our approach from experimentation to predictable success: Automated Iterative Context Building: Building context progressively rather than overwhelming LLMs with data upfront Standardized Prompt Architecture: Creating structured templates that simplify implementation across use cases Optimized Data Presentation: Formatting data to significantly improve accuracy and reduce hallucinations I'll demonstrate these through our "Kubernetes Deployment Failure Investigation" system, where our LLM approach outperformed traditional coding by considering a wider range of factors. Attendees will gain practical frameworks for systematic LLM application development applicable within and beyond Kubernetes, with opportunities to share experiences in this rapidly evolving field.
Andrey is the Open Source Dev Leader at Komodor, a startup building the next-gen troubleshooting platform for Kubernetes. With over 20 years of experience in the dev space, Andrey is known as the founder of JMeter-Plugins.org, an open source project which he led to its undisputed status as the premier source of JMeter plugins in the world. Heâs also an avid speaker who loves talking about Open Source (could you have guessed?!), performance testing, all things automation, and his favorite programming languages Python, Golang and Java. Prior to Komodor, Andrey was the Co-founder of Up9 and before that the Open Source Evangelist at BlazeMeter.com.
