Informatics graduate from Universitas Gunadarma, class of 2026. The thesis built a hybrid Autoencoder + CNN that sorts MRI scans into four tumour classes. Internal test accuracy reached 89.31 percent on 4,735 cleaned images. On BraTS 2020 data the model had never seen, 71.25 percent. Most of the work happens before training. Auditing and merging three public MRI sources moved accuracy further than any architecture change did. The pattern repeated in the web and mobile projects: the unglamorous step decides the result. The rest is delivery. Python and Flask on the server. TypeScript and React Native on the client. Turso and SQL Server in the data layer. A model in a notebook is still a hypothesis. What gets built around it decides whether anyone can trust the output.

Keahlian teknis