Projects

Toward Engineering AGI: Benchmarking the Engineering Design Capabilities of LLMs

Research Paper

Conducted comprehensive benchmarking of Large Language Models' engineering design capabilities, evaluating their performance across various engineering design tasks and methodologies.

Skills: Python, Signal Processing, Computer Vision

YOLO Object Detector for PASCAL VOC

Ayush Barik, Christopher Kim

ECE494 Class Project

Implemented the YOLOv1 object detection algorithm from scratch. The project's core was developing the intricate YOLO loss function to simultaneously optimize for bounding box regression, objectness confidence, and class probabilities. Trained on the PASCAL VOC 2007 dataset, the model achieved a mean Average Precision (mAP) of over 0.5.

Skills: Python, PyTorch, Computer Vision, Object Detection, Deep Learning

Denoising Diffusion Models and Rectified Flow

Ayush Barik, Christopher Kim

ECE494 Class Project

Developed a UNet-based diffusion model from scratch, progressively adding time and class conditioning. Trained the model for conditional image generation on both MNIST and ImageNette. Also implemented and benchmarked Rectified Flow against the standard DDPM, comparing performance and inference speed.

Skills: Python, PyTorch, Deep Learning, Diffusion Models, UNet, Rectified Flow, Computer Vision

Real-Time QVGA Video Streaming with Filters on FPGA

Joanna Li, Ayush Barik

ECE385 Class Project

Engineered embedded quantized live video streaming pipeline on a Spartan-7 FPGA to transmit data from an OV7670 camera, optimized for low power (0.35W) and minimal resources (813 LUTs, 28 BRAM, 309 Flip-Flops). Implemented additional video filters and on-screen sprites.

Skills: SystemVerilog, Python