About me
I’m a first-year computer science student at BINUS University in Indonesia. I like building small, specific things for people around me: a translator for Manado Malay, and two web tools for classmates.
At a glance
- Based in
- Indonesia
- Studying
- Computer Science, BINUS University (first year)
- Interested in
- NLP and machine learning, especially language models for low-resource languages
- Tinkering with
- Hardware electronics
- Projects
- Three, all on the Work page
- Currently building
- MangARTI mini, the same translator on models of around one billion parameters
- Languages
- English, Indonesian
Where it started
Most of what I’ve built started with something small and specific: classmates missing the shuttle, students drilling the same math concept without making progress, a dialect that mainstream models have barely seen. None of them are big problems, but they were close enough for me to notice.
The question I ask about all of them came from a peer’s question about my final high school project, which turned MangARTI into a translator. Now I ask it first: who will use this, and what will they do with it?
What I’ve picked up so far
Most of it comes from the translator. Fixing its hallucinations taught me to check the data before I touch any training settings. Native speakers wrote the translations in the dataset, and they’re still the people who can tell me when an output sounds wrong. Lately I’ve also started scoring every model on the same hand-written test set, so I can compare runs fairly.
The smaller projects taught me something different: BSQShuttle and IterMath are static pages with no model and no server, and they didn’t need either.
What I’m working on now
MangARTI mini: the same translator on Qwen and Llama models of around one billion parameters, so it can run on ordinary hardware. The first three runs, and how they compare with the 8B model, are on the Work page.
If you try one of the projects and something breaks or sounds wrong, I’d like to know.