Real-time road damage detection from a moving vehicle, optimized for edge devices. YOLO11m compressed with mixed precision quantization (INT8 + FP16, TensorRT): 1.92–2.97× faster inference and ~65% smaller model while staying within ~1% of FP32 accuracy. Includes a live dashboard with GPS-tagged detections, map view, and damage-area estimation.
Web-based mental health support application built as a Coding Camp 2026 capstone. Led a 6-person team from concept to deployed product; developed and deployed an IndoBERT-based emotion classification model reaching 86% accuracy on Indonesian text, handling the full ML workflow from preprocessing to integration.
IoT monitoring system for public street lighting deployed at Simpang 5 Keputih, Surabaya. An ESP32 acquires data from five environmental sensors over RS485 and streams readings to a cloud REST API for a real-time dashboard. Covered full field implementation: sensor communication setup, pole installation, hardware troubleshooting, and calibration.
Image classification model for the Animals-10 dataset built with MobileNetV2 on TensorFlow/Keras, using a custom Sequential architecture with Conv2D and MaxPooling2D layers.