Edward Wibowo

Software teams need real-time task tracking without complex setup or loose permission management across project members.

Built a modular REST API using FastAPI and SQLModel with custom middleware for role-based authorization (owner/member levels) and Dockerized deployment.

Delivered a robust backend API with relational PostgreSQL schemas, achieving low-latency query execution and containerized delivery on Render.

Mobile transaction fraud requires real-time detection on heavily imbalanced datasets, where standard accuracy metrics fail to capture high-risk fraudulent behavior.

Engineered an end-to-end inference pipeline featuring dynamic transaction feature generation (rolling stats & balance shifts), threshold-tuned Random Forest evaluation, and Flask REST API integration with MongoDB persistence.

Processed and prepared over 1.67M dataset rows, achieving ~97.3% fraud recall on test set evaluations while establishing a sub-second model deployment workflow on Vercel.

Disaster assessment requires rapid, accurate flood severity categorization from visual data, where standard classification ignores the continuous, ordered nature of flood progression.

Built an end-to-end computer vision workflow in PyTorch, fine-tuning an ImageNet-pretrained EfficientNet-B0 with image augmentation, ordinal regression loss formulation, and structured train/val/test evaluation.

Successfully engineered dataset preprocessing and multi-class evaluation routines, validating model performance using Mean Absolute Error (MAE), confusion matrices, and detailed error distribution reviews.