iFood MLOps App

This repository is my solution for the iFood ML Engineer test.

TL;DR

git clone https://github.com/andre-marcos-perez/ifood-mlops-app.git
cd ifood-mlops-app
docker-compose up

1. Getting Started

1.1. Introduction

The iFood MLOps App is a ML platform built on top of Docker designed to help data scientists to develop and deploy ML models. It is composed by four main components tied together by a Python SDK:

• Sandbox: Jupyter based env to develop models and interact with the platform;
• Pipeline: Apache Airflow based pipeline to train, test and deploy models;
• Database: MySQL based database to register ML experiments and predictions;
• Serving: FastAPI based API to serve predictions.

The ML platform dynamics is best understood with this example notebook that will be available at runtime.

1.2. Pre-requisites

Install Docker (v1.13.0+) and Docker Compose (v1.10.0+).

2. Architecture

2.1. Applications

App	Address	Description
Sandbox	localhost:8888	Jupyter based env to develop, train, test and deploy models
Pipeline	localhost:8080	Apache Airflow based pipeline to train, test and deploy models
Database	localhost:3306	MySQL based database to register ML experiments
Serving API	localhost:8000	FastAPI based api to serve models predictions and predictions
Serving Docs	localhost:8000/docs	Swagger based web docs for serving api

2.2. Description

For a further description of each application, please checkout docs.

3. Build

3.1. Local build

You can build the project locally.

Note: Local build is currently only supported on Linux OS distributions.

git clone https://github.com/andre-marcos-perez/ifood-mlops-app.git
cd ifood-mlops-app
./build.sh

3.2. Remote build (recommended)

Alternatively, you can use the images pre-built on DockerHub. The images are built with GitHub actions, you can check the code here.

4. Run

4.1. Run local build

If you build the project locally, you can start the platform with the following command:

docker-compose -f docker-compose-local.yml up

4.2. Run remote build (recommended)

If you prefer to use the pre-built images, start the platform by running the following command:

git clone https://github.com/andre-marcos-perez/ifood-mlops-app.git
cd ifood-mlops-app
docker-compose up