hx f

I’m an AI and machine learning researcher who builds across generations of models—from spiking neural networks to modern LLM agent systems. I focus on practical training strategies while keeping a benchmark-driven mindset.

In my graduate research, I worked on spiking neural networks (SNNs) and deep learning methods, including DLa and reinforcement learning components. My work emphasized rigorous experimentation (SNN training, architecture choices, and metric improvements) to make models more reliable in real settings.

More recently, I explored LLM-based agent pipelines using PyTorch, Python, and methods like RAG, SFT, and DPO. I’ve also implemented VLM-related work and evaluated across benchmarks, reporting meaningful improvements (e.g., jumping from 65% to 80% in one evaluation) and strengthening query/retrieval and preference-learning workflows.

I enjoy turning complex ideas into working systems—clean ablations, clear metrics, and fast iteration—so the research can move from experiments to dependable performance.