In our previous article, we wrote about an end-to-end blood cell classifier for cancer prediction, a final project for ML Zoomcamp created by graduate Alexander Daniel Rios. In this follow-up, Alexander reflects on how the course shifted his approach from ad-hoc experimentation to a structured, professional workflow. He walks through the skills he developed, the two final projects he completed, and why working in public and with a community made a difference for his professional growth.
ML Zoomcamp is a free, four-month online course on machine learning fundamentals and deploying models to production.
What were your projects like before ML Zoomcamp?
Alexander: Chaotic. I learned many concepts on my own in a disorganized way. My notebooks were hard to maintain, I didn’t follow a clear structure, and I often jumped from one model to another without proper validation or traceability.
What changed during the course?
Alexander: I learned to articulate my thoughts more clearly and work in a professional, end-to-end manner. The recorded lessons, hands-on assignments, and final projects helped me structure my workflow and effectively utilize the correct tools, from development to deployment.
Which skills did you develop?
Alexander: I learned a lot of practical skills! Some of them:
- Systematic data preparation
- Applying regression and classification with proper validation
- Selecting appropriate evaluation metrics
- Importantly, understanding when and why to use a specific model.
I also gained confidence with neural networks, transfer learning, and making inferences with models deployed via Flask or AWS Lambda.
What projects did you work on?
Alexander: We were encouraged to do two final projects and participate in a Kaggle competition. My first project was about creating a classifier to predict cancer in blood cells. I talked about it in the recent article on the DataTalksClub blog.
My second project focuses on classifying disaster-related tweets. I also wrote an article about this project, in which I explore how I utilized spaCy, TensorFlow, and BERT-based architectures for NLP.
I found both projects engaging because they address real-world challenges. Their complexity let me apply what I learned and push further with concepts beyond the core curriculum.
And what about your Kaggle competition experience?
Alexander: As part of the course, I joined the ML Zoomcamp 2024 Retail Forecasting Competition on Kaggle. The task was to predict next-month product demand using 25 months of historical data across multiple stores. I applied feature engineering, time series analysis, model evaluation, and deployment. The pipeline included temporal feature extraction, economic indicators, and robust validation.
The final model achieved 4th place globally, among dozens of participants. This strengthened my skills in real-world forecasting, time-series modeling, efficient preprocessing, and competition-style thinking.
How did the course contribute to your capstone work?
Alexander: The ML Zoomcamp course provided the theoretical and practical foundations I needed. It offered a structured path from ML fundamentals to pipeline design and deployment. I gained hands-on experience preparing and validating datasets, selecting models, and evaluating them properly. I also worked with production tools such as Docker, Flask, and TensorFlow Serving, which were key to transforming the project into a functional product.
What kept you motivated?
Alexander: The active community, practical challenges, and guided projects encouraged me to take the work beyond an academic setting. Although the final project was the starting point, the methods and mindset I learned helped me scale the solution, add components like segmentation and multitask learning, and build a robust deployment.
How did sharing your work affect your growth?
Alexander: A lot. Each week, I posted on LinkedIn about the tasks I was working on, the challenges I faced, and what I was discovering. That habit led to connections, feedback, and visibility, resulting in conversations with people from diverse backgrounds, suggestions to improve my work, and even job offers and freelance requests. As a natural extension, I created my first personal blog focused on ML Zoomcamp, where I wrote weekly reflections, explained my decisions, and shared what I learned.
This experience helped me grow technically and as a communicator, and it gave me the confidence to share my work beyond the classroom. It showed the value of learning in public and being part of a community.
Thanks for sharing your experience with us, Alexander!
Conclusion
Alexander’s journey started with scattered notebooks and unclear validation and ended with a reproducible workflow, well-chosen metrics, and deployable models. ML Zoomcamp provided the structure and the practical stack to make that shift possible. Along the way, community and public sharing turned weekly progress into feedback, opportunities, and confidence.