Insights into the Future of Serverless Computing, Generative AI, and MLOps

Stockholm, Sweden – The annual AWS Summit 2023 united tech enthusiasts and industry leaders from across the Nordics to delve into the latest trends and innovations in cloud computing. Boasting a fully packed agenda featuring notable speakers and esteemed vendors from the likes of Amazon Web Services, Novo Nordic, and Snowflake, attendees were treated to profound insights into the forefront of cloud offerings and real-world applications.

Some of my favorite insights were:

Generative AI

The past year has been a thrilling journey for generative AI, a fact not lost on AWS, which has recently introduced Amazon Bedrock in preview on its platform. Amazon Bedrock is a fully managed service that grants access to foundational models from leading AI startups as well as Amazon itself through a convenient API. Personally, one of the most crucial aspects of Amazon Bedrock is its ability to generate a private copy of the foundational model, accessible exclusively to the customer, which is then trained within the customer’s Virtual Private Cloud. This approach ensures that data never leave the confines of the customer’s private sphere, maintaining utmost privacy as many companies require. 

MLOps

Among the key highlights of the conference, Niklas Palm’s presentation on MLOps in SageMaker stood out to me. Palm powerfully explained how data scientists have transitioned from the mere concerns of “how to train and deploy a model” to the far more nuanced considerations of “how to re-train, monitor, and reproduce results of a model.” The resounding answer lies in the realm of MLOps, or rather, improved MLOps. Palm went on to explain how AWS Sagemaker addresses these challenges through an array of features, including model versioning, automated code- and model pipelines, and scalable infrastructure for both training and deployment.

I, too, recognize the pattern described in his observation that data science teams have matured significantly. One aspect that I consistently observe as noticeably absent in real-world scenarios is the limited application of DevOps principles to machine learning projects, which hampers the goal of accelerating the development and deployment of ML models. This crucial aspect of the model lifecycle has gained substantial traction, and it was uplifting to witness its formalization within AWS Sagemaker.