2026-09-09 –, Ballroom All times in US/Pacific
What does it actually take to run LLMs in production and why is it so hard to find clear answers?
This talk is a practical walkthrough of the core MLOps concepts behind modern LLM systems, built from scratch out of necessity. Faced with a scattered ecosystem and limited end-to-end guidance, I set out to piece together how everything fits: from GPUs and custom silicon, to ML pipelines and fine-tuning, to the realities of serving models at scale.
In 30 minutes, I will try to lay out the fundamentals, such as hardware tradeoffs, model optimization techniques like quantization, and the mechanics of high-performance inference such as KV caching, batching, and distributed serving.
This isn’t a polished framework or step-by-step guide, but it is a cohesive map of the basics. If you’ve struggled to understand how to “DevOps” LLMs, this talk will help you bridge the gaps and build your own mental model.
Maarika Teose is a technology leader with over a decade of experience driving innovation across diverse industries. She is currently the Head of DevSecOps for a global automotive OEM, where she focuses on delivering secure, scalable software development platforms for next-generation Advanced Driver Assistance Systems. Previously, she led DevOps for Powell’s Books, and was a Release Manager at Jama Software. She is passionate about leadership and developing high performing teams. She is a proud mom of two kiddos and loves to dance.