A complete archive of my work in research, robotics, and applied deep learning.
Logit-Lens interpretability for VLMs. Prompt-free transformer classifier over VLM logits — 83% on DARPA dataset.
ViT-from-scratch on CIFAR-10 (76.5%). CLS-token 6D satellite pose via ViT-B transfer. Jetson Orin + TensorRT @ 17ms.
BERT-based web firewall trained on 2M safe HTTP requests · 98% accuracy. Novel KD-LoRA continual learning. 45ms inference.
100K-image synthetic spacecraft pose dataset spanning orbital geometry and illumination conditions for ISRO-sponsored research.
Self-refining VLM trauma detection pipeline on UAV embedded hardware. ~80% accuracy on DARPA emergency response dataset.
Decoding-time alignment for T2I diffusion: timestep-aware localized inpainting + REINFORCE policy over CFG, noise jitter, DDIM inversion.