An undergraduate specializing in Deep Learning and Machine Learning. Building scalable training pipelines and exploring open-world perception through transformer-based vision models.
Hello! I'm Dhruv, a researcher and developer passionate about demystifying black-box AI. Currently pursuing my B.Tech in Computer Engineering at Vidyalankar Institute of Technology (2023 - 2027).
I thrive on understanding how neural architectures behave under constraints. Whether I am experimenting with Vision Transformers from scratch, analyzing bipartite matching loss in object detection, or scaling models on parallel data pipelines, I enjoy tackling complex architectural problems.
Designed a YOLOv8-based model to detect trash disposal actions in real-time. Deployed model with TensorRT on Jetson Nano, achieving 22 FPS inference speed for smart city waste monitoring applications.
Trained an EffNetB2 + scratch ViT model on 20% of the Food101 dataset (101 classes). Designed scalable training with regularization and tuned LR schedules, achieving 97.22% accuracy via ViT.
Recreated the Vision Transformer architecture from scratch based on the original research paper. Implemented patch embedding, multi-head self-attention, and positional encoding to study scaling laws and convergence.
Used MediaPipe for real-time skeleton keypoint extraction during batting motions. Developed form evaluation algorithms achieving 92% consistency with expert scoring. Built a Streamlit dashboard for playback-based biomechanical feedback.
Designed and evaluated transformer-based vision models for object detection and segmentation in simulated environments. Experimented with DINOv2 and Grounding DINO.
Completed industry-driven ML certification. Built supervised ML models achieving 95%+ accuracy and automated ML workflows using Python and scikit-learn.
I am actively looking for internship and research opportunities relating to computer vision and deep learning starting in 2026. My inbox is always open.