Experiments
Things that started as a question.
Half-finished ideas, research notebooks, and weekend builds. Kept here because deleting them felt dishonest.
Rodent detection dataset
Building the training set behind SATRIA: public rodent data merged with night footage shot in Sedayu, Grobogan, annotated for rat, mouse, and the false positives that cost the most.
Indonesian sentiment analysis
Fine-tuning IndoBERT on Indonesian text, and finding out how much of sentiment is sarcasm the model has no idea about.