Steven is a qualified Data Scientist and Data Engineer who develops and deploys full analysis pipelines for Machine Learning and data mining projects on large datasets. He has hands-on experience translating models into business insights and applying data knowledge to real-world business problems. Steven uses various technology stacks to deliver solutions to complex projects using a practical focus and collaborating in cross-functional teams.
Creating and implementing data analysis pipelines including data access, ingestion, munging/manipulation/cleansing, analysis/modeling, testing, and deployment/integration into business applications and services.
Enhancing operational aspects of businesses by increasing control of company data.
Working in cross-functional teams to provide data-driven solutions for increased efficiency and productivity.
Organized and analyzed large amounts of data by developing analysis pipelines and performed visualization of results via dashboards for non-technical audiences.
Assisted the business decision-making process by implementing adaptive Machine Learning systems on large time series datasets.
Implemented software in a Scrum team and collaborated with engineering and product development teams to meet customer requirements.
Managed project activities in a goal-oriented manner, evaluating the knowledge transfer between domains of artificial and real images to minimize costs for labeled real data.
Enhanced Deep Learning approaches using Python to extrapolate patterns from large sets and predict new data by analyzing available datasets.
Created Machine Learning tools that analyzed front car camera images to highlight objects with bounding boxes.
Deep learning methods have achieved state-of-the-art results in Computer Vision tasks like image classification, object detection, and face recognition.
The project focused on optimizing the use of artificial data for training deep neural networks, specifically addressing the need for labeled data and the transferability of learned models to real-world data.
A two-stage fine-tuning process was implemented, starting with pre-trained weights and gradually incorporating mixed training data to reduce the required target domain data by approximately half.
