The Fun-Sized MLOps Stack from Scratch
Companies & organizations know they shouldn’t build for Google but they also don’t know how NOT to build for Google scale. A majority of documented stacks have been by companies that reached data and ML maturity years ago or by bleeding edge startups. These stacks are either fully-constrained on their toolchain or have been created for incredibly specialized use cases. The MLOps tooling ecosystem is fragmented and companies that are just starting on their journeys to becoming ML-native or ML-fluent are confused by the ML Ops maturity models that don't account for their particular organizational goals or trajectory, especially if they're not "on the road" to Google maturity. My goal is to show "fun-sized companies" (SMB's, small startups, etc) how to build a fully fledged MLOps platform from scratch using the best OSS tools out there in under a day. Specifically we'll cover: What are the main problems MlOps tries to solve What are the most common tools being used & their drawbacks What are some OSS projects & tools that have been developed in the past 2-3 years and how do they solve some of the pain points of the prior tools What is the realistic roadmap for companies that are forever “not-Google” scale but want to continue improving their data and ML maturity