This project is an edge-deployed computer vision and anomaly detection system engineered for high-risk industrial facilities to autonomously identify hazardous hydrocarbon gas leaks and optical plume dispersals in real time.

Focusing on machine learning to detect gas anomalies and autonomous gas leaks directly from camera feeds, I engineered a TensorRT-optimized vision pipeline deploying YOLOv8, SAM2, and RT-DETR alongside specialized thermal architectures (GasSeg, LangGas, CGNet) on Jetson Orin Nano edge-servers, fusing legacy point telemetry with PTZ-mounted MWIR optical gas imaging cameras.

This edge architecture sustained 24/7 plume segmentation at 30 FPS across 3+ industrial monitoring sites without thermal throttling, delivering a spatial tracking dashboard that slashed false-positive alerts by 45%, reduced leak identification latency to under 120 milliseconds, and benchmarked hardware configurations over 20+ runs to eliminate integration errors.

Conducted offshore formation evaluation and reservoir characterization for Anambas exploration wells alongside PT LAPI ITB, Pertamina, BP, KUFPEC, and Halliburton, screening proprietary logging workflows across 10+ client deliverables to de-risk multi-million-dollar energy investments. Automated Python and Julia tensor pipelines to impute missing well logs, analyze core scanner data (LIBS, XRF, hyperspectral, RGB core photo), and compute water saturation in complex LRLC reservoirs via hardcoded Modified Simandoux and Thomas-Stieber algorithms, accelerating multidisciplinary interpretation timelines by 45%.

In joint research with ITB Reservoir Engineering, co-invented a patent-pending Python/C++ framework using OpenCV to extract geological textures from RGB core imagery, boosting resistivity prediction accuracy to 90% (+5 point gain) and cutting turnaround from 3 days to 1 day. Coupled Physics-Informed ChebyKAN neural networks with UA-ORST spatial tracking to predict drilling parameters and shear-sonic logs (DTS) with strict rock-physics compliance, reaching 0.895 R² on DTS benchmarks, limiting underbalance violations to 3.7%, and reducing raw acquisition overhead by 37% across 4 multi-well domains.

Investigated sub-picosecond carrier dynamics and spin-phonon coupling in Si(001), bulk tungsten, and emergent altermagnets (MnF₂) under Prof. Changyong Song and Jungchan Choi at POSTECH.

Executed 28-femtosecond pump-probe spectroscopy (800 nm) and THz-TDS measurements, successfully extracting a 32 fs momentum and 260 fs electron-phonon relaxation time.

Engineered GPU-accelerated computational imaging pipelines in Python, C++, and CUDA using extended Ptychographic Iterative Engines (ePPM) to reconstruct high-fidelity diffraction images from massive synchrotron datasets.