I’m a data scientist with 4 years of experience working on real-world machine learning and analytics projects. I enjoy building end-to-end data products and sharing what I learn along the way.
Nothing here yet.
Up to the previous post, the project was in a reasonably good state: data scraping worked, feature engineering was stable, models were trained and selected via MLflow, and offline inference behaved as expected. The system seemed to do what it was sup...

Phase 4A was relatively smooth: I built a production-ready inference pipeline with modular preprocessing, caching, consistent schemas, and an MLflow-tracked model. Phase 4B was supposed to be the easy part: containerize the pipeline, wrap it in a Lam...

In real-world ML, an accurate model is only the starting point. The real challenge is delivering reliable predictions at scale, ensuring that every step at inference, preprocessing, encoding, and feature alignment mirrors training. In this phase, I b...

In real estate, extreme listings, both ultra-luxury and severely underpriced, can disproportionately influence predictions. These outliers, though rare, carry meaningful signals that must be preserved while keeping our model stable and robust. In thi...
