My Projects
TDOA Geolocation
As a Signals Analysis Intern at Zeta Associates, I developed a Python and X-Midas program for TDOA geolocation using signals from 4 drones. I was responsible for coding and transmitting an encrypted BPSK signal and implemented parallelized cross-ambiguity functions for faster-than-real-time TDOA value calculation.
AI Pothole Detection
For my senior design project, I was an AI Model and Data Developer on a team creating an app to live-identify potholes. I personally curated a dataset of 22,239 images and used YOLOv8 to train and implement the model, achieving a 0.656 mAP50. Our project was awarded best in computer science out of 33 teams.
License Plate Identification
I developed an AI model to first localize license plates in images and then identify the digits using EasyOCR. For this, I built a neural network in Python using NumPy and CV2, training it on a dataset of 2131 images.
OpenFHE Code Generation
To address the challenges new researchers faced with CKKS, I developed an LLM-powered agent to generate efficient and accurate OpenFHE code. This agent handled tasks like addition, multiplication, dot product, matrix multiplication, and convolutions from natural language inputs. The project incorporated inference-time reasoning optimizations, including specific prompting, decoding, self-improvement, and Retrieval-Augmented Generation (RAG) techniques. I evaluated its performance using various LLMs and metrics like Pass@k (for compilability and functionality) and CrystalBLEU, aiming to simplify FHE development.