Chandru-21/MLOps_Project

An end-to-end MLOps pipeline(CI/CD/CT/CM) project for training, versioning, deploying, and monitoring machine learning models using FastAPI, Kubernetes, MLflow, DVC, Prometheus, and Grafana.

Machine Learning Opearations (MLOps)

MLOps maturity level 4

Overview :

This project implements a robust MLOps pipeline, facilitating the continuous integration, continuous deployment, and monitoring of machine learning models. The infrastructure leverages AWS, Kubernetes, and various open-source tools to ensure scalability, reproducibility, and maintainability.

Architecture :

Key Features :

Data Versioning : DVC

Continuous Integration(CI) : Triggered through ‘main.yml’ , building the code (docker), tests the code(Pytest),pushes the docker image to AWS ECR.

Experiment Tracking / Model Versioning : MLflow

Continuous Deployment(CD) : Deploys FastAPI in AWS EKS(kubernetes cluster) for real-time and batch predictions.

Continuous Monitoring(CM) : Integrating the ‘/metrics’ method of FastAPI in Prometheus and visualizes endpoints in Grafana.

Continuous Training(CT) : Triggers code execution through GitHub Actions when new data is pushed to the remote DVC location and committed to Git.

Drift Monitoring : Uses a Streamlit app to monitor data drift, target drift, and performs data quality checks.

Data Monitoring :

Data Drift Monitoring :

Data Quality checks :

Continuous Monitoring(CM)

Exposing "/metrics" on FastAPI to be connected to Prometheus :

FastAPI integrated in to Prometheus :

Integrating Prometheus in to Grafana for Visualization :

Monitoring FastAPI methods on Grafana,

Monitoring the resources of the Kubernetes cluster on Grafana :