Welcome to Course 2 of the Practical Data Science Specialization. In this course, you'll learn how to build an end-to-end machine learning pipeline, in the cloud. You build on the topics covered in Course 1, and you create a text classifier for product reviews, using the BERT family of NLP models. BERT transformer network language model caused quite a stir in the machine learning community, by showing that it could achieve state-of-the-art results in many different NLP tasks, such as text classification and name entity recognition. I hope you find deploying BERT in the cloud to be fun. I'm thrilled again to be here with your instructors for this course, Sireesha Muppala, Shelbee Eigenbrode, and Antje Barth. Sireesha, perhaps you could say a bit more about what's coming up in this course. Sure, thank you Andrew. In Course 1 of the specialization, you worked on building a text classifier model for product reviews, you exploded the data set, and created models, using both AutoML and the Amazon Sagemaker placing text built in algorithms. In this course, you will build a custom model using the BERT algorithm and build an end-to-end machine learning pipeline for your product review classified. BERT, as Andrew just mentioned, is a state-of-the-art natural language processing algorithm, based on transformer architecture. BERT's key innovation is to apply the bidirectional nature of the transformer architecture to language modelling. Previous efforts in language modelling looked at a sentence sequence in one direction, either from left to right or right to left. In contrast, BERT supports simultaneous bidirectional training, leading to very impressive results. In the first week of this course, I will show you how to generate machine learning features from the raw data and share those features with other teams, using a scalable feature store. Now Antje will take over, to dive into details of Week 2, which describes how to use these features to train a model, at scale, in the cloud. That's right. In week 2, you will dive into the bird model architecture and learn how to build and train your custom bird model, which is usually a two step process. The first step is to pre-train your model in an unsupervised learning step, and the second step is then to fine tune the model for a specific language task. The exciting thing here is that there are many pre-trained bird models available, that you can start using with a simple Python API call. You will get hands-on experience on how to fine tune a pre-trained bird model for the specific task of multi-class classification for product reviews. You will also learn how to monitor and profile the model training in this fine tuning step, which gives you the visibility and the control needed to continuously improve your models. Once you have the trade model, Shelbee will show you how to automate this process through an end-to-end machine learning pipeline in week three. Indeed, and this is where it gets really interesting. In the third week of the course, you'll learn how to orchestrate the various steps of feature engineering, model training, model evaluation, and model deployment, into an end-to- end machine learning pipeline. You'll also explore pipelines in the context of the broader field of machine learning operations, or ML apps for short, and focus on automating that end-to- end machine learning workflow. Finally, you'll learn about end-to-end traceability, by tracking artifacts and model lineage through the pipeline. That sounds exciting! Being able to use your local machine or a local jupiter notebook to do things like feature transformation, model training, and testing is great. But as you see in this course, there's a lot more involved in deploying an end-to- end machine learning pipeline in the cloud. Learning to operationalize your machine learning models is key to ensuring that the business outcomes you hope to achieve are actually achieved by your data science initiatives. So it is very exciting to see ML apps also covered in this course. With that, let's dive into building, training, and deploying machine learning pipelines using BERT.