Summary of "Week 6 Lab: AI-ML Services in AWS and Azure"

Week 6 Lab: AI/ML Services in AWS and Azure

This lab focuses on cloud-based AI/ML services in AWS and Azure, with emphasis on MLOps — especially the deployment phase (taking a trained model to production).
You should already know basic ML (regression/classification) and have example code ready (can be taken from ChatGPT or other sources) so you can focus on cloud deployment rather than ML theory.
Many cloud ML services are paid even on student/free tiers — always check pricing before using a service.
Both platforms provide notebook environments similar to Jupyter/Colab for developing, training, evaluating, and deploying models.
There are many pre-trained (managed) AI services (e.g., image recognition, text-to-speech) you can call without training your own models.
Monitoring is important: use CloudWatch (AWS) or equivalent tools to observe application behavior.

Emphasis: deploying and exposing your model as an API/endpoint is central to productizing ML.

Storage
- Upload dataset to an S3 bucket.
Data preparation / ETL
- Use AWS Glue or AWS Data Wrangler to clean and transform data. (Glue is emphasized.)
Notebook & training
- Use Amazon SageMaker (notebook instance similar to Jupyter/Colab) to write and run Python/ML code, load modules, train models, and evaluate metrics.
Model deployment
- Create a SageMaker endpoint for real-time inference. The endpoint exposes an API — send input and receive predictions.
Monitoring
- Use Amazon CloudWatch to monitor logs and performance (applies to ML endpoints, Lambda functions, web apps).
Suggested practice
- Complete at least two projects: one basic and one more advanced.
- Focus especially on learning Glue and SageMaker endpoints.

Storage
- Upload data to Azure Blob Storage.
Data preparation / ETL
- Use Azure Data Factory or Azure Databricks to prepare and visualize data.
Notebook & training
- Use Azure ML Studio (workspace / ML studio notebooks, similar to Colab/Jupyter) to write and run training code.
Model management
- Register the trained model in Azure ML (use the model registry).
Model deployment
- Deploy the model as an endpoint (gets a URL you can call with inputs to receive predictions).
Monitoring and testing
- Call the endpoint URL to provide input and test predictions.

Prioritize deployment (MLOps): exposing a model as an API/endpoint is key to production value.
Use pre-trained / managed AI services when appropriate (image recognition, speech/text tools) to save time and cost.
Always verify whether cloud services are free or chargeable; if chargeable, avoid running them unnecessarily or document costs.
CloudWatch (AWS) is a general monitoring tool useful across serverless, web, and ML workloads.
SageMaker and Azure ML Studio behave like Jupyter/Colab environments — the notebook-based development experience is similar across platforms.

AWS: S3, AWS Glue, AWS Data Wrangler, Amazon SageMaker, SageMaker endpoints, Amazon CloudWatch, AWS Lambda
Azure: Blob Storage, Azure Data Factory, Azure Databricks, Azure ML Studio (ML workspace)
Notebook/tools: Jupyter, Google Colab, Anaconda
External helper: ChatGPT (as a source for sample code)
Pre-trained services (general category): image recognition, text-to-speech, etc.