About me

Hi, I’m Nevin, a Master’s in Data Science student at the University of Wisconsin - Madison. I’m passionate about transforming data into meaningful insights—whether through analytics, engineering robust data pipelines, or building intelligent machine learning systems. With expertise in Python, SQL, cloud technologies, and machine learning, I enjoy solving complex problems and optimizing data-driven solutions.

Beyond data, I’m an adventurer and storyteller. I run a YouTube channel where I share my travels and I also love capturing the world through photography. Whether it's developing AI models or framing the perfect shot, I enjoy the process of discovery and innovation. If you share my enthusiasm for AI, data science, or even just a great adventure story, let’s connect!

Ask NevBot anything about me!

Experience

Experience

  1. Data Science Intern Wisconsin School of Business – Madison, WI Sept 2024 — Present

    • - Investigated Starbucks unionization impact using GPT-3.5-Turbo and fine-tuned BERT on customer reviews, finding a 12% rise in negative sentiment post-unionization.
    • - Streamlined an SQL-driven ETL pipeline by automating data preprocessing and transformation, cutting manual intervention by 90% for seamless fine-tuning of BERT.
    • - Optimized GPT API costs by 25% compared to the baseline prompt strategy through prompt optimization and rate-limiting strategies while ensuring high-quality sentiment analysis.
    • - Designed an Etsy artwork price prediction model with CLIP-based multimodal embeddings, improving accuracy by 23% over embeddings calculated by ResNet50 model and OpenAI API.

  2. Assistant in AI/Data Research UW College of Agricultural & Life Sciences – Madison, WI Sept 2024 — Mar 2024

    • - Engineered cranberry phenology classification model using ResNet50, improving accuracy by 25% over baseline CNN model. Implemented semi-supervised learning with CLIP model for auto-labeling 12K+ images.
    • - Architected an object detection model with YOLOv8 to identify cranberries by color variation, enhancing precision by 15% through custom data augmentation (Albumentations) for better fruit selection.
    • - Established end-to-end ML pipelines on AWS (S3, SageMaker) integrated with MLflow, enabling seamless model versioning and deployment.
    • - Collaborated with cranberry growers to understand their needs, ensuring the models provided actionable insights for optimal fertilization timing and efficient fruit selection.

  3. Graduate Researcher Wisconsin Institute for Discovery – Madison, WI Jan 2024 — Jun 2024

    • - Implemented an automated active learning pipeline integrating CLIP and Stable Diffusion (SD), reducing manual annotation by 80% while improving labeling consistency across ImageNet, CIFAR-10, and CIFAR-100 datasets.
    • - Innovated an image labeling system, where SD refined CLIP's predictions, ensuring high-confidence labels while minimizing human intervention and reducing annotation errors by 30% compared to a semi-supervised approach.
    • - Benchmarked model performance against 20+ research papers, demonstrating superior labeling accuracy over existing methods, validating its effectiveness for large-scale dataset annotation.

Contact

Contact Form