Adopting MLOps is getting increasingly important with the rise of AI. A lot of different features are required to do MLOps in large organizations. In the past, you had to implement these features yourself. Luckily, the MLOps space is getting more mature, and end-to-end platforms like Databricks provide most of the features. In this talk, I will walk through the MLOps components and how you can simplify your processes using Databricks. Audio for this session is delivered in the conference mobile app, you must bring your own headphones to listen.

In the world of GenAI, advancements are happening at a crazy speed. These advancements concern not only the algorithms but also the operations side of things. In this talk, we will go back to the basics, discuss the main principles of building robust ML systems (traceability, reproducibility, and monitoring), and explain what types of tools are required to support these principles for different types of applications.

We talked about:

Maria's background Marvelous MLOps Maria's definition of MLOps Alternate team setups without a central MLOps team Pragmatic vs non-pragmatic MLOps Must-have ML tools (categories) Maturity assessment What to start with in MLOps Standardized MLOps Convincing DevOps to implement Understanding what the tools are used for instead of knowing all the tools Maria's next project plans Is LLM Ops a thing? What Ahold Delhaize does Resource recommendations to learn more about MLOps The importance of data engineering knowledge for ML engineers

Links:

LinkedIn: https://www.linkedin.com/company/marvelous-mlops/

Website: https://marvelousmlops.substack.com/

Free MLOps course: https://github.com/DataTalksClub/mlops-zoomcamp Join DataTalks.Club: https://datatalks.club/slack.html Our events: https://datatalks.club/events.html

At Ahold Delhaize, we have 19 local brands. Most of our brands have common goals, such as providing personalized offers to their customers, a better search engine on e-commerce websites, and forecasting models to reduce food waste and ensure availability. As a central team, our goal is to standardize the way of working across all of these brands, including the deployment of machine learning models. To this end, we have adopted Databricks as our standard platform for our batch inference models.

However, API deployment for real time inference models remained challenging due to the varying capabilities of our brands. Our attempts to standardize API deployments with different tools failed due to complexity of our organization. Fortunately, Databricks has recently introduced a new feature: serverless API deployment. Since all our brands already use Databricks, this feature was easy to adopt. It allows us to easily reuse API deployment across all of our brands, significantly reducing time to market (from 6-12 months to one month), increasing efficiency, and reducing the costs. In this session, you will see the solution architecture, sample use case specifically used to cross-sell model deployed to four different brands, and API deployment using Databricks Serverless API with custom model.

Talk by: Maria Vechtomova and Basak Eskili

Connect with us: Website: https://databricks.com Twitter: https://twitter.com/databricks LinkedIn: https://www.linkedin.com/company/databricks Instagram: https://www.instagram.com/databricksinc Facebook: https://www.facebook.com/databricksinc