18 hours ago
Remote, CanadaSenior
Responsibilities
- Design, train, and ship recommendation and ranking models, including session-based embeddings, collaborative filtering, content embeddings, and visual embeddings.
- Build and operate low-latency Python model-serving services for live storefront traffic.
- Run daily batch pipelines that rebuild model artifacts across hundreds of merchants.
- Design and analyze A/B tests and make launch decisions based on evidence.
- Own production deployments, including containers, Kubernetes manifests, autoscaling, dashboards, and alerts.
- Extend the LLM shopping assistant and LLM-assisted catalog-enrichment systems with appropriate evaluation.
- Operate streaming systems and maintain production machine-learning infrastructure.
Requirements
- At least 5 years of experience building machine-learning systems, including at least 3 years serving models to live production traffic.
- Strong Python experience with FastAPI or an equivalent framework.
- Experience with vector databases and approximate-nearest-neighbor libraries such as Qdrant, FAISS, Annoy, ScaNN, or HNSWlib.
- Hands-on recommender-systems or search-ranking experience involving implicit feedback, embeddings, and approximate-nearest-neighbor search.
- Strong SQL skills against analytical stores; ClickHouse experience is a plus.
- Comfort with Docker and Kubernetes and willingness to own deployments.
- Experience running A/B tests and reporting unfavorable results objectively.
- Experience with streaming systems such as Pulsar, Kafka, or Flink.
- Bonus experience with PyTorch, sentence-transformers, CLIP, computer vision for product imagery, gradient-boosting rankers, scikit-learn, scipy, LLM application evaluation, tool calling, or e-commerce and search relevance.
- Python asynchronous programming experience and experience around search.
Tech Stack
Apache FlinkApache KafkaCircleCIClickHouseDockerFastAPIGoogle Cloud PlatformGrafanaHugging Face TransformersKubernetesLightGBMMySQLNumPyPandasPostgreSQLPythonPyTorchRedisscikit-learnSciPySQLXGBoost
Categories
About Maropost
Maropost builds a unified commerce platform for ecommerce merchants and retailers, combining online storefronts, merchandising, marketing automation, customer service, and payments in one SaaS suite. Founded in 2011 and headquartered in Toronto, it is privately held and serves thousands of brands, including Mercedes‑Benz, Seiko, and Fujifilm. The company has been recognized on Deloitte’s Technology Fast 500 and G2 category lists.