greggompers

Packaged Generative AI chatbot models on AWS Bedrock such as LLama3-70B and Anthropic Claude 3 Sonnet, Haiku LLMs in combination with Langchain embedding models into an internal PIP Python package for use across the organization. Rebuilt Production ML pipelines responsible for $30M in annual revenue into AWS architecture. Most recent pipeline executed an ensemble of 9 PyTorch models, outputting 63 million rows of predictions with 800GB+ compute memory capacity through containerization on Amazon SageMaker Pipelines. Tools and technologies used included PyTorch, Tensorflow, and Amazon SageMaker - directly applicable to real-time trading signal generation and high-throughput execution systems. Enhanced ML pipelines with 5,000 to 15,000 lines of code to include automatic model drift monitoring on a daily/weekly/monthly basis. Implemented using Python, Amazon SageMaker Pipelines, PySpark, AWS Glue, CloudWatch, and IAM roles. Published 20 videos as training material with 2,500 internal views from Machine Learning Engineers and Data Scientists. Focused on topics including machine learning model development, AWS SageMaker, PyTorch, TensorFlow, and best practices in MLOps pipeline construction.

Freelance with Multiple Clients. 100% satisfaction rating with 5-star reviews for all 4 solutions delivered to clients. Managed and delivered projects on Upwork, using languages and tools such as Python, SQL, Pandas, and Scikit-learn. Utilized project management tools like Trello and communication tools like Slack and Zoom to ensure client satisfaction and timely delivery. Feature engineering on TechCrunch financial datasets of 7,000,000 rows using PySpark. Used domain expertise to create a target variable and add multiple columns to the dataset to capture hidden metrics. Achieved 99% accuracy in a multi-class classification pipeline using XGBoost.

Predicting Startup Investment Decisions. Achieved 60% accuracy in predicting the growth of new startups over the next 12 months using a fine-tuned XGBoost model with custom feature engineering on AWS. Employed tools like Jupyter Notebooks for development and experimentation, and Amazon S3 for data storage and management. Engineered features on AWS SageMaker to extract target labels and generate new features from a raw dataset of 57 features and 6000 samples.