research

My research focuses on representation learning for complex time-series data, specifically electrocardiogram signals. I explore multiple self-supervised techniques, such as contrastive learning frameworks and joint-embedding predictive architectures, to extract meaningful features from biological data. My goal is to create rich and meaningful embeddings from vast amounts of unlabeled data, which can be used as a foundation for downstream classification of cardiac anomalies.

I also do research on TinyML. Currently, I'm addressing the challenges of processing high-resolution data for defect detection on edge devices, where memory and compute are extremely constrained. Overcoming these hardware constraints is crucial for low-latency applications, particularly in fields such as drone imaging and retail logistics.

I'm interested in expanding current AI capabilities and research methodologies. I'm actively exploring RL techniques to enhance LLM reasoning. Additionally, as a hobby, I build agentic tools that empower researchers and streamline research workflows.