I am a Master's student at Texas A&M University bridging the gap between scalable systems engineering and interpretable AI. My work focuses on Large Language Models (LLMs), Chain-of-Thought reasoning, and multimodal representation learning.
Proposed a memory-efficient framework (AmCLR/xAmCLR) for bimodal contrastive learning. Achieved superior retrieval performance on CC3M using significantly smaller batch sizes (128) compared to CLIP baselines.
Developed a mechanistic interpretability pipeline to identify reasoning-critical neurons in LLMs. Enhanced these neurons to achieve a 14% accuracy boost on the GSM8K benchmark.
Engineered a dual-model regularization framework to fine-tune SDXL using LoRA and SaRA. Improved style alignment by 19% (CLIP Score) while preventing catastrophic forgetting.
Implemented a minimal x86 kernel in C++ featuring a complete virtual memory subsystem, recursive page table mapping, and a preemptive multi-threaded scheduler.