Katonic MLOps Certification Course

Introduce myself and share twelve years of experience across data roles and product management, including startup work building government machine learning products in Odisha, and scaling machine learning applications.

Explore Docker and Kubernetes with practical examples, running a simple Python Flask app and examining environment differences across Linux, Windows, and macOS.

Explore containers as portable wrappers that package code, dependencies, and configuration, with Docker enabling isolated Python environments (TensorFlow, other frameworks) on any host.

Explore virtual machines with dedicated RAM and CPU under a hypervisor and containers that share the host OS, highlighting their lightweight nature and rapid creation with live demos.

Learn to create and run Docker containers by writing a Dockerfile, building a Docker image with Python packages from requirements, exposing a port, and launching a container with an entrypoint.

Discover how Kubernetes manages containerized applications with Docker, enabling automatic scaling, health checks, and rollback across multiple machines in cloud or on-prem environments, via load-balanced services.

Explore pods as the basic unit in container orchestration, and compare vertical and horizontal scaling with containers, replicas, and YAML deployment.

Learn how Kubernetes deployments enable auto scalability and fault tolerance by configuring replicas, containers, and images, then expose services with load balancers and ingress routing.

Learn how namespaces create logical groups to isolate services and deployments across multiple environments, enabling you to manage dev, qa, and test inside a single Kubernetes cluster.

Follow a walkthrough that converts a Flask app into a Docker image, runs it locally, and then deploys to Kubernetes with deployments, services, and port mappings.

Explore the full MLOps stack from data gathering to deployment, including pipelines, source control, experimentation, hyperparameter tuning, scalable training, canary deployment, model registry, feature store, monitoring, and governance with Kubernetes.

Explore the evolving MLOps landscape, where no single platform fits all; cloud providers, open source integrations, and diverse competitors require you to weave together multiple services.

Explore Katonic MLOps platform, a unified tool suite with ui, monitoring, and development layers that lets data scientists use their language and algorithms while accelerating experimentation, deployment, security, and governance.

Turn raw data into meaningful information by crafting features to predict the probability of a customer leaving, exploring contextual features, augmentation, and pre computed features.

Learn to convert experimental data science workflows into scalable ai model lifecycle pipelines, from data exploration and feature engineering to model registry, deployment, monitoring, and creating new model versions.

Explore the Katonic MLOps platform by browsing diverse use cases, building scalable pipelines from notebooks to production, and managing experiments, deployments, and monitoring in one integrated workspace.

Meghna introduces herself as the instructor, a data scientist with about five years of experience, and presents a demo of the platform.

Learn how to operationalize AI applications on a unified platform, from data preparation to deployment, including workspace creation, grid integration, pipelines, and model deployment.

Extract movie genres from the movie overview using image and text processing to power a personalized recommendation system on online platforms.

Explore the AI model lifecycle from data cleaning and feature transformations to experiment-driven model selection, versioning, and deployment, and learn automated monitoring and retraining for continuous improvement.

Explore the platform by selecting a popular use case, launch notebooks and a deployed app to build a text classification pipeline that predicts a movie's genre and tone.

Create and name a workspace in the platform to enable experimentation in labs or Jupiter notebooks, then select a Jupiter Lab environment and a memory image for movie review demo.

Connect to the augmented workspace, switch to the correct one, and fetch data from a report using the terminal. Check out the required branch and access the notebooks.

Explore a prediction notebook that fetches IMDb overview data from the movie database API via an access token, preprocesses text, builds bag-of-words features, and trains tuned classifiers with 80/20 split.

Explore the experiments tab to view versioned trials, logged measures and parameters, and select the best barometers and model using available metrics.

Learn how to register a model, verify if it's already registered, and create a new model. Implement the model registration pipeline to ensure proper workflow.

Learn to build and deploy an ml pipeline with git-based versioning from github, step dependencies, docker images, and private bucket storage to enable parallel training and reproducible runs.

Deploy a model by promoting a selected version to production, archive previous versions, configure deployment with classification or regression settings, and use secured API endpoints and time-limited tokens for predictions.

Create a workspace and build a Streamlit app to visualize model results for stakeholders, then deploy it with one click and secure access with cluster credentials.

Set up an automated inference pipeline in production back-end that fetches daily data, transforms and cleans it, runs deployed models, and generates continuous predictions with visualization for monitoring.

Learn how to schedule a pipeline run for inference, selecting versions and experiments, and configure daily or weekly executions with api tokens and private bucket for daily data and predictions.

Monitor deployment performance by tracking inference rate, latency, memory, and request counts, using the feedback dashboard to detect issues and identify outliers, and trigger retraining when performance drops.

Retrain models on a monthly schedule, monitor drift, and automatically trigger the training pipeline to deliver fresh models as part of the end-to-end lifecycle.