Skip to content

0.1. Projects

What is the default learning project?

This course's default project is a forecasting task using the Bike Sharing Demand dataset. Your goal is to predict bike rental demand based on factors like weather and time. A reference implementation is available if you need guidance.

Forecasting is an ideal starting point for MLOps because it mirrors many real-world business problems and introduces common challenges. You will learn to handle time-series data, prevent data leakage by correctly partitioning data, and build a robust pipeline from feature engineering to model deployment. This project provides a solid foundation in MLOps principles that is transferable to other domains.

Can I use my own project instead?

Yes, you are highly encouraged to use a personal or professional project. Applying the course concepts to a domain you already understand can accelerate your learning and deliver immediate value.

Working on your own project allows you to: - Directly improve a system you are passionate about. - Avoid the learning curve of a new dataset and problem space. - Build a portfolio piece that is uniquely yours.

How can I find good project ideas?

If you're looking for inspiration, data science competition platforms are excellent sources for well-defined problems with ready-to-use datasets:

  • Kaggle: A premier platform for data science competitions, datasets, and community collaboration.
  • DrivenData: Focuses on social impact projects, allowing you to use your skills for good.
  • Hugging Face Datasets: Offers thousands of datasets, ideal for projects in NLP, computer vision, and audio.

What makes a project suitable for this MLOps course?

Whether you choose the default project or your own, a good project for learning MLOps should have:

  • A Clear Objective: A well-defined goal, such as improving a business metric or solving a specific problem.
  • Accessible Data: Data that is readily available and sufficient for training a model.
  • Deployment Potential: The model should be something you can imagine deploying, even if in a simulated environment.
  • Measurable Outcomes: Clear metrics to evaluate model performance and project success.

Can I work on a Large Language Model (LLM) project?

While this course focuses on traditional predictive ML, the MLOps principles you'll learn—like versioning, automation, and monitoring—are foundational and applicable to Large Language Model (LLM) projects.

However, Generative AI introduces unique challenges not covered in depth here, such as:

  • Specialized Infrastructure: Requiring powerful GPUs for fine-tuning and inference.
  • Complex Evaluation: Moving beyond simple accuracy to metrics like BLEU/ROUGE or even LLM-based evaluations.
  • New Techniques: Involving prompt engineering, retrieval-augmented generation (RAG), and vector databases.

For these reasons, we recommend starting with a predictive ML project to master core MLOps fundamentals before tackling the complexities of LLM Operations (LLMOps).