8 hours ago
Responsibilities
- Own the complete production lifecycle of ranking and recommendation models, including validation, deployment, versioning, retraining, rollback, and retirement.
- Design scalable ML data and feature engineering pipelines using Spark/PySpark and lakehouse technologies, including feature generation, aggregations, and training dataset construction.
- Build monitoring for data quality, model drift, prediction quality, service reliability, latency, and infrastructure cost.
- Develop reusable training, evaluation, deployment, orchestration, and model versioning infrastructure for multiple models and teams.
- Create a model hub containing documentation, evaluation results, release history, versioning, and monitoring information.
- Establish release criteria, rollback paths, CI/CD workflows, alerts, incident-response procedures, and operational runbooks.
- Partner with data scientists and engineering teams to define evaluation signals, preserve scientific validity, set MLOps priorities, and improve operational reliability.
- Mentor teammates and establish reusable engineering standards across the ML portfolio.
Requirements
- 5+ years of experience building and operating production ML systems beyond initial deployment, including monitoring, retraining, incident response, and continuous improvement.
- Strong production software development skills in Python and proficiency with SQL.
- Hands-on experience with Spark/PySpark for large-scale distributed data processing and feature engineering.
- Experience with modern data lake or lakehouse architectures, including S3, Apache Iceberg, or equivalent technologies.
- Experience with workflow orchestration for scheduling, dependency management, retries, backfills, and recovery.
- Hands-on experience building infrastructure with Terraform and operating containerized workloads on Kubernetes.
- Experience building reusable training, evaluation, or deployment infrastructure used by multiple models, scientists, or teams.
- Experience with model and dataset versioning, experiment tracking, lineage, reproducible releases, CI/CD, monitoring, alerting, rollback, and recovery procedures.
- Ability to distinguish infrastructure failures from changes in model effectiveness and balance model quality, delivery speed, reliability, and infrastructure cost.
- Strong cross-functional communication and experience partnering with data scientists.
- Preferred: experience with Go, model registries or model hubs, delayed outcome labels, configurable ML infrastructure, cross-team adoption of standards, mentoring, and access controls for sensitive information.
Categories
About Fetch
Fetch builds a consumer rewards app where users earn points by scanning receipts and redeem them for gift cards, while brands use its SKU-level, retail-agnostic purchase data for targeted, outcomes-based advertising and measurement. The privately held company, founded in 2013, processes over 13 million receipts daily, creating visibility into more than $200 billion in gross merchandise value. Its platform has awarded over $1 billion in points and operates a hybrid-remote workplace.