I've built machine learning projects spanning Indonesian food-image analysis and Steam review sentiment analysis, using Python, PyTorch, Scikit-learn, and Tableau.
For an Indonesian Food Ingredient & Nutrition Analyzer, I developed an end-to-end multimodal pipeline combining ResNet18, web-mined weak supervision, and Qwen2-VL prompting. My few-shot VLM achieved over 6x higher micro-F1 than the CNN baseline, and I conducted error analysis on domain shift, out-of-vocabulary ingredients, false predictions, and hallucinations.
On a Steam review analysis project, I cleaned and manually labeled 171 reviews, then built a TF-IDF and Logistic Regression classifier that reached 86.7% accuracy and 86.4% macro-F1. I also created a Tableau dashboard to surface ambiguous feedback and recurring player concerns.
Alongside my data science work, I've supported programming and communication courses at Universitas Indonesia and coordinated mentoring operations for COMPFEST's Data Science Academy. I enjoy helping learners apply programming, debugging, structured communication, and computational problem-solving skills.
