Developed a multimodal image retrieval system using FAISS, CLIP-ViT (Vision Transformer), and Flask, enabling image and text-based queries to improve search efficiency in large unstructured collections.

Designed a data extraction pipeline from Mapillary and fine-tuned YOLOv8 and YOLOv9 models to detect CCTV cameras, achieving F1 scores of 68% and 81% on test sets to analyze their utilization, deployment, distribution, and privacy impact.

Extracted and transformed data to identify racial discrepancies in sentencing within counties of California, employing advanced analytics to uncover potential biases.

Designed an automated pipeline for a Yale University lab to extract and summarize court cases, enhancing accessibility through Flan-T5 and GPT-4 integration.

Researched and applied the Google Operations Research Tools VRP Solver to enhance logistics operations for John Deere.

Tailored a solution with multiple capacities and resources, time windows, multiple visits, ordered visits, vehicle reloading, and other constraints, resulting in a 36% decrease in depreciation expense per productive hour.

Predicted candidate galaxies for deeper observation of Low Surface Brightness Structure (LSB) using galaxy simulations.

Implemented a CNN in Keras/TensorFlow, achieving 83% accuracy in classifying low surface brightness galaxies.