George Welson

• Adapted adiffusion model (Generative AI) for generating handwritten Arabic text by implementing a positional encoding step to handle the 2-4 distinct shapes of each character, using Tensorflow and CUDA. • Addressed data scarcity in the ADAB dataset with a multithreaded augmentation script, ensuring each character had at least 2,500 samples. Evaluated the model’s performance using Fr´echet Inception Distance (FID) and Geometry Score (GS).

• Developed a real-time copper price prediction model using machine learning (PyTorch) and deployed it with Streamlit for interactive forecasting and visualization, acheiving 93% accuracy. • Developed and deployed a smart sales analytics chatbot powered by GPT API and Google Cloud Vertex API, LangChain and Retrieval-Augmented Generation (RAG), enabling natural language access to structured sales data and boosting stakeholder insights through conversational analytics.

• Completed Information Storage and Management v5 Associate course. • Worked on investigating AI bias using adversarial debiasing and SMOTE.

• Developed and implemented a YOLOv8-based object detection system, achieving 96% accuracy for object recognition. • Designed a photogrammetry model to accurately estimate object distance and direction, enhancing the vehicle’s situational awareness. • Contributed to the team’s victory, securing 1st place in the Robotaxi competition by delivering a robust and reliable perception system.

• Developed an interactive sales data dashboard, using PowerBI, that improved data visibility over a map and contributed to securing 2 new client contracts. • Built and deployed a robust ETL data pipeline using Python and SQL, streamlining data extraction, transformation, and loading for analytics workflows (from MongoDB to PostgreSQL)

• Developed and implemented an Ant-Colony Optimization algorithm to determine the most efficient shortest path for the autonomous vehicle. • Deployed the entire path planning and decision-making system onto the Robot Operating System (ROS) framework for real-world application.

Exchange Semester

Data Science Track