Join our Data Science Enablement squad as a Senior Machine Learning Engineer. You will use an existing batch inference model to establish a secure, automated deployment pipeline. This role involves both engineering and change management, including architecture and training, with a focus on educating data scientists and other Data Science Enablement members on MLOps. Once the foundational deployment framework is in place, you will enable additional MLOps capabilities such as MLFlow, A/B testing, real-time endpoints, and further automation with Model Risk Management (MRM).
Key Responsibilities:
- Develop and implement a secure, automated deployment pipeline.
- Educate and mentor team members on MLOps practices.
- Balance engineering tasks with change management and training.
- Enhance MLOps capabilities with advanced tools and techniques.
Preferred Experience:
- Experience in highly regulated industries like banking, finance, or healthcare.
Qualifications:
- Experience:
- Minimum of 3-5+ years of experience in machine learning and MLOps.
- Proven experience with AWS Sagemaker and building end-to-end machine learning models.
- Experience with data integration and management using IBM DB2 and Snowflake (or like databases)
- Strong understanding of CI/CD pipelines and automation tools.
- Technical Skills:
- Proficiency in programming languages such as Python, R, SQL and/or Java.
- Use of Fifth Third standard DevOps tools such as Jira, Terraform, GitHub, Jenkins
- Knowledge of containerization and orchestration tools (e.g., Docker, Kubernetes).
Squad outcomes:
- Future (2025 & Beyond) – Utilize AWS Sagemaker to expand Feature Store, introduce Model Registry, CI/CD, Real-Time models for our large data science credit models.
- The squad is currently working on an in-house build of Feature Store to help speed up modeling process for our Data Science department. Combination of Snowflake, Cloud Pak for Data. (More on this later)
- Currently, data scientist build model features (attributes) about customers in their own Jupyter notebook that feed into their models and never reuseable for others… aka reason for Feature Store
- They are also working on building real time scoring framework for our loan/card application process. Right now it’s batch and can be almost 31 days behind.
- Technology used: Docker, Kafka, Snowflake, Feature Store