Introduction 🧠

You’ve trained a model. It works great on your laptop. You ship it. Six months later, nobody’s maintained it, the predictions are garbage, and your data scientist has moved on. Sound familiar?

Every ML team at some point. Don’t be this dog. 🐶🔥

That’s exactly the problem this project tackles — head on, in three progressive phases. We’re building a customer churn prediction system for retail, starting from a messy Jupyter notebook and ending with an autonomous AI agent that monitors drift and retrains the model without you lifting a finger.

Introduction

Here we are at the end of our long journey in the cloud which led us to implement our serveless system in synchronous and asynchronous mode.

Our application meets the basic functional requirements to be used in production, however we cannot release a digital system without worrying about security. Security allows us to retain our customers, to assure them of the protection and anonymity of their data which passes through our application. It also protects us against malicious attacks.

Introduction

In our previous article, we focused on the lambda-Sync part of our serverless system which corresponds to the synchronous part of our application.

We had already set up our entire continuous integration pipeline consisting of Github for the VCS, Codebuild to build our docker images and finally Cloudformation to deploy our resources in the AWS environment.

For this second part, there is no need to reinvent the wheel. We will use the instruments already in place and implement our asynchronous circuit.