Major Projects
Autonomous Drone with a Compact and Coaxial Redundant Motor System for Disaster Relief Using Spatial AI
A ready prototype of a compact autonomous drone leveraging Spatial AI in the form of depth mapping for the aid in disaster relief, rescue and recovery. This design features a custom 3D printed compact frame comparable to a F450 in size, featuring a coaxial X8 motor configuration for motor redundancy in case of a motor failure.
Dockerization of Location-based Panoptic Segmentation Model
Containerized a location-based panoptic segmentation model built on the EfficientPS architecture with SOLOv2 as the instance segmentation head. Quantized model weights to INT8 using TensorRT for efficient deployment on edge devices, combining advanced segmentation techniques with resource-efficient deployment strategies.
Smart Microwave System Simulation (Embedded Systems Intern)
Designed and simulated a smart microwave oven in PICSimLab, implementing algorithms to calculate optimal cooking times based on food parameters. Developed a user-friendly interface and ensured functionality through rigorous testing. Utilized MPLAB IDE, XC8 compiler, and C programming for implementation, building on foundational skills in microcontrollers and SDLC-based embedded systems development.
X-Ray Crystallography Based Epitope Mapping of Glycoproteins and RNA in Chandipura Vesiculovirus for Vaccine Design
Conducted an independent research study on epitope mapping of the Chandipura Vesiculovirus utilizing its 3D structure, a virus with a high mortality rate of up to 75%. Analyzed Glycoproteins and RNA Polymerase L to identify highly immunogenic peptides for virus neutralization. Utilized computational tools like ChimeraX and PyMOL alongside databases such as Protein Data Bank (PDB) and National Library of Medicine (NLM).
Residual Graph Convolutional Network for Protein Structure Understanding with Self-Derived Spatial Features
Developed ResidualGNNPlus, a novel Graph Convolutional Network that automatically reconstructs 39 biophysical features - including hydrophobicity, solvent accessibility, and secondary structure - directly from 3D protein structures. Using a residual GCN architecture with layer normalization, the model achieves R² > 0.9 on large proteins and operates efficiently on standard hardware, enabling accessible protein analysis for drug discovery and bioinformatics.