All case studies
Master of SWE & AI · Machine Learning·2026
Review Pulse — multi-domain sentiment classifier
An ML pipeline that classifies product-review sentiment and holds up across four retail domains.
AI/MLData
Problem
Product reviews carry signal that’s expensive to read at scale, and a model trained on one category often degrades on another. The task: build a sentiment classifier that generalises across multiple, different retail domains — not just one.
Approach
- Worked from a labelled dataset of ~8,000 Amazon reviews across four domains (Books, DVDs, Electronics, Kitchen & Housewares).
- Cleaned and vectorised the text, then trained and evaluated classical ML models with scikit-learn, following a CRISP-DM workflow.
- Measured per-domain and cross-domain performance to test generalisation — not just headline accuracy.
- Iterated in Jupyter notebooks, version-controlled in a dedicated open-source repo.
Stack
Pythonscikit-learnpandasJupyterCRISP-DM
Outcome
- A working multi-domain sentiment classifier with documented per-domain and cross-domain evaluation.
- A reproducible ML pipeline (notebooks + repo) — concrete, hands-on ML, not just coursework slides.