Rise of Serverless Databases

Something About Databases // Alex DeBrie

Having an informative chit-chat about Serverless and DynamoDB with Alex DeBrie was completely enlightening. He is the Founder of DeBrie's Advisory, a consulting company that mainly focuses on AWS technologies and databases.

The Serverless Rabbit Hole
Serverless is a managed event-driven function-based computing. Its architecture defines an entry point for the code that will be triggered whenever an event occurs, such as messages or streams.

The asynchronous nature of event-driven architectures makes it challenging to share schemas, especially when they evolve during development. With a structured pattern, it will be easier to understand the data schemas and data when sharing schemas between decoupled serverless producers and consumers.

However, serverless architecture scales better and is more beneficial due to the decoupling, especially when building machine learning and data engineering pipelines.

Database Land
The use cases of databases within the cloud ecosystem have matured from pulling traditional databases like MySQL and Postgres into cloud data centers to ones that are elastic enough to use different primitives of the cloud. The early stages of database infrastructure that was built for the cloud were either Online Transactional processing (OLTP) based like DynamoDB or Online analytical processing (OLAP) based like Snowflake. But as each day goes, we get new databases with new tricks.

When evaluating a database, it's more beneficial to pick one with a very opinionated structure and a specific use case rather than one with a checklist of features and capabilities as the value prop.

Video || Spotify || Apple

Designing Machine Learning Systems by Chip Huyen.

Machine Learning Systems Overview

"Machine learning systems" are frequently thought of as simply the ML algorithms that are utilized, such as logistic regression or various forms of neural networks.

Unfortunately, the algorithm is only a minor element of a production-ready ML system. The system also includes the business requirements that spawned the ML project in the first place, the interface through which users and developers interact with your system, the data stack, and the logic for developing, monitoring, and updating your models, as well as the infrastructure that allows that logic to be delivered.

The Relationship Between MLOps and ML Systems Design

Ops in MLOps is derived from DevOps, which stands for Development and Operations. The term "operationalize" refers to putting anything into production, which includes deployment, monitoring, and maintainence. MLOps is a set of tools and best practices for putting machine learning into production.

The design of ML systems uses a system approach to MLOps, which means that it evaluates an ML system holistically to guarantee that all of its components and stakeholders can collaborate to meet the set objectives and criteria.

Read Here

MLOps for Large Natural Language Models

At the Tel Aviv MLOps meetup, Elhay Efrat, a Cloud Solution Architect for startups, walks attendees through the practical application of Open AI on Azure.

Elhay delves into the nitty-gritty of how to leverage this powerful platform to enhance machine learning performance and efficiency in real-world scenarios.The session provides an overview of the machine learning lifecycle in real-world applications, including a comparison of cognitive services versus Open AI. It covers ML operations on Azure, and how to use Open AI to take their machine learning to the next level in terms of performance and efficiency.

YouTube