At Lingnan University, I fine-tuned Google’s TimesFM for industrial time-series forecasting, reducing MAPE by approximately 11% versus an LSTM baseline across three held-out datasets. I also designed monitoring pipelines that improved early fault detection sensitivity by approximately 12% over SPC baselines.
I built Dockerized FastAPI inference services and deployed machine-learning workflows on AWS EC2 for anomaly scoring and model serving.
During my Ph.D. research at Indian Institute of Technology Delhi, I developed GAWNO, a wavelet neural operator architecture for fault detection and isolation. It improved FDI accuracy by 8–14% against the baselines I evaluated and was published in Control Engineering Practice.
I also developed probabilistic and multimode monitoring methods for process and nuclear systems, including work with real NPCIL sensor data. My research has appeared in journals including Process Safety and Environmental Protection and Progress in Nuclear Energy.

