// ml_engineer · deep_learning_researcher

>_

M.Sc Informatics @ Institut Teknologi Sepuluh Nopember

Machine Learning Engineer translating mathematical frameworks into deployable AI systems. Current research in temporal graph learning, medical image segmentation, and imbalanced classification.

Graph Neural Networks Medical Image Segmentation Imbalanced Classification Computer Vision MLOps

education

Master's in Informatics

Institut Teknologi Sepuluh Nopember

2025 — Present

Research on Explainable Temporal GNNs for financial forensics and Deep Cross-Level Context Fusion for medical image segmentation.

gpa3.68
Data Mining Applied Data Science Computer Vision Big Data

Bachelor's in Information Technology

Institut Teknologi Sepuluh Nopember

2020 — 2024
gpa3.47
Artificial Intelligence Big Data Management Data Structures Data Modeling

Machine Learning Path

Bangkit Academy — Led by Google

2023
TensorFlow Data & Deployment Data Analytics Math for ML

research & engineering

Evaluated systems — metrics from real benchmarks, not demos.

Explainable Graph Attention Network (X-GAT) — Bitcoin AML Detection

2026

Explainable GNN (Temporal GATv2 + GNNExplainer) using GraphSMOTE and Time2Vec to detect illicit transactions on the Elliptic dataset under strict chronological splits. Mitigates severe temporal concept drift.

illicit f10.751 precision0.839 pr-auc0.774
PyTorch PyG XGBoost GNNExplainer

CCFNet+ — Cross-Level Context Fusion for Polyp Segmentation

2026

CCFNet+ (Res2Net-101) with Coordinate Attention and adaptive context gating to resolve extreme scale variations and boundary ambiguity in medical polyp segmentation.

mean dice0.905 mean iou0.849 s_α0.931
PyTorch OpenCV Res2Net Albumentations

University Demand Forecasting System

2025

Hybrid time-series system combining Facebook Prophet (logistic growth) with heuristic fallback to predict elective course demand under volatile post-pandemic trends. Solves the cold-start problem for new courses.

allocation accuracy90.6%
Prophet Pandas Time-Series

Diabetes Detection with Cost-Sensitive Learning

2025

Classification pipeline handling 1:3 class imbalance via Weighted Random Forest and multi-modal feature integration (clinical + sociobehavioral). Proves algorithmic cost-sensitive learning beats SMOTE on data integrity.

roc-auc89.3% recall85.4%
Random Forest SMOTE/ADASYN Scikit-learn

live demos

Deployed models — interactive, hosted on Hugging Face Spaces.

Image classifier built and trained with TensorFlow and the EfficientNetB1 architecture.

TensorFlow Keras EfficientNetB1 FastAPI

experience

DevOps Engineer Intern

PT MYECO Teknologi Nusantara

Sep 2023 — Dec 2023
  • Secured 100% of web traffic by configuring an Nginx reverse proxy within a CI/CD pipeline with SSL on port 443.
  • Cut average application response time by 40% by establishing a Redis caching layer.