Deep dives of the week
Generative AI keeps rolling out with breath-taking advancements. Last week, Google Cloud unveiled a slew of generative AI tools for developers for application development, ML modeling, and more. Meanwhile, Microsoft continues to up the ante: Azure introduced new VMs for generative AI that exa-scale LLMs by 30x over the previous generation.
In this week's special package of NPW Insights, we're showcasing four diverse developments from Google, Netflix, Azure, and Shopify that demonstrate the wide-ranging innovation driven by AI/ML today.
Our other spotlight this week is on serverless technology, which has been one of the most consequential technologies in cloud computing lately. But with the enterprise-ification of this technology, says an experienced cloud architect, new challenges are emerging. A must-read musing.
Finally, don't miss the fascinating story of an engineering team that took on the unconventional task of migrating from Lambda to Kubernetes, how they successfully bridged the gap between the two technologies for a smooth transition.
Insights from Netflix’ media understanding platform
To help creatives perform dialogue, visual, and reverse-shot search on its content library without having to watch it, Netflix engineers first developed a tightly coupled application, where ML algorithms were tied to backend and UX code. This impeded scalability and development of new use cases, which led to the development of a media understanding platform – which serves as an abstraction layer between apps and ML algorithms. It lets apps interact via a GraphQL or gRPC interface, which forwards queries to a Search Gateway. The latter modifies queries to match the target dataset, and maps it to multiple searchers which then execute it. STORY
Learnings from a reverse migration: from Lambda to K8s
Liav Yona, Engineering Team Lead from Firefly, a multi-tenant IaC creation service, explains the core idea that made Lambda to Kubernetes migration seamless: “Eventually, what’s a lambda? It’s a type of job that needs to be done a single time with a specific configuration that runs a bunch of workers to get the job done. This brought us to the epiphany that this sounds a lot like…K8s jobs.” STORY