Completed Projects

Conducted a comparative study on DCNN, DarkNet-19, ResNet-50, and Efficient B0 architectures, achieving a 94.35% validation accuracy with Efficient B0 on a dataset of over 3500 32x32 grayscale images across 28 classes. Custom-designed DCNN achieved 92.35% validation accuracy with the lowest training time, demonstrating efficient model design.

Implemented music signal recreation using a trained CNN model. Leveraged genre-specific feature representations synthesized into new music signals. Features were converted into audio waveforms, demonstrating the potential of neural networks to generate realistic genre-specific samples. Validated quality through reclassification and performed PCA analysis to examine variance distributions across four genres.

Co-optimized dataflow and hardware architecture for DL models (VGG16 and UNet) using the MAESTRO cost model. Focused on minimizing latency and energy consumption while adhering to predefined hardware and dataflow constraints. Demonstrates expertise in the co-design space of DL accelerators comprising deep learning models, dataflow, and hardware architecture.

Developed a comprehensive autonomous driving stack in ROS2, integrating perception, planning, and control modules. The system enables real-time navigation and decision-making in dynamic environments, demonstrating end-to-end autonomous vehicle capabilities from sensor data processing to vehicle control.

Despite advancements in automotive technology, traffic accidents remain a significant global issue. This project creates and simulates various traffic scenarios using the CARLA simulator to understand driver behavior. Data gathering includes vehicle metrics and biometrics such as heart rate and breathing rate to enhance safety through better understanding of driver behavior patterns.

Integrated and prototyped the datapath and control units of a simple 32-bit MIPS processor featuring five pipeline stages using Verilog hardware description language. The processor executes various instructions including arithmetic, logic, data movement, and flow control operations with optimized pipeline hazard handling.

Designed and developed a high-performance edge detection algorithm module using SystemC for image and video processing at 24fps. The system demonstrates real-time processing capabilities suitable for embedded vision applications.

Focused on enhancing data transmission efficiency within communication networks by implementing Parallel to Serial and Serial to Parallel Interfaces. Explored the integration of these interfaces with the OSI model's seven layers, demonstrating understanding of layered communication protocols.

Created a safe and intelligent control environment ensuring the safety of stingrays, reptiles, and elephants in zoo settings. This project leverages embedded systems, cloud-based programming, and a software-based user interface to monitor, collect, and store environmental data in real-time.

Implemented a BREACH attack to explore its potential impact on targeted systems. Researched and selected appropriate attack methods, designed and executed the attack on simulated systems using Python, and analyzed results through HTTPS requests between client and server to understand vulnerability exploitation.

Presents the design of a human-aware automotive testbed using CARLA (Car Learning to Act), incorporating human-vehicle interaction and environment sensing to enhance testing and validation of autonomous systems. The testbed enables realistic vehicle behavior simulation including traffic, pedestrians, and environmental conditions, ensuring robust system performance in various scenarios.

Space missions increasingly rely on nuclear power, notably Radioisotope Thermoelectric Generators (RTGs) and nuclear fission reactors. This paper reviews their applications, focusing on Nuclear Thermal Propulsion (NTP), analysis, and data acquisition. RTGs provide reliable energy for space probes and rovers. Nuclear fission reactors offer potential for high-energy propulsion, enabling faster space exploration.