Machine Learning Engineer

Role Overview

We are hiring a Machine Learning Engineer to build production ranking, recommendation, and personalization systems: the models that decide which options a user is shown, in what order, and why. This is a classical ML role at its core, covering feature engineering, gradient-boosted and linear models, learning-to-rank, retrieval and search relevance, and the classification and propensity models that sit alongside them. You will own problems end to end, from framing an ambiguous product question as an ML problem through to a deployed, monitored model, and success will be measured on demonstrated online impact rather than offline metrics alone.

Key Responsibilities

• Build and productionize ranking and recommendation models that select and order the best options for a given user request, balancing relevance to the user against business objectives.
• Frame ambiguous product problems as ML problems: define the target, the labelling strategy, the training data, and the offline metric that credibly predicts online impact.
• Design and build feature pipelines over behavioral, transactional, and content data, with point-in-time correctness so training features match what is available at inference and no label leaks into the model.
• Develop and tune classical ML models (gradient-boosted trees, linear and logistic models, learning-to-rank), and recognize when a simpler model or a rules baseline is the right answer.
• Build multi-stage retrieval and ranking architectures: candidate generation, eligibility and business-rule filtering, scoring, and re-ranking, all within a production latency budget.
• Improve search and match relevance using lexical and embedding-based retrieval, hybrid approaches, and re-ranking, with query understanding where it moves the metric.
• Build classification, segmentation, and propensity models used to score and route incoming demand.
• Handle the constraints real ranking systems have: sparse and skewed data, cold start for new users and new inventory, position and selection bias, feedback loops, and eligibility or capacity limits on what can be recommended.
• Own offline evaluation: temporally honest splits, ranking metrics such as NDCG, MAP, and recall@k, calibration, and slice-level analysis to find where a model fails rather than reporting a single average.
• Design and analyze online experiments, including A/B test design, power analysis, guardrail metrics, and clear-eyed interpretation of results, including negative ones.
• Deploy models to production as low-latency inference services or batch scoring jobs, with versioning, reproducibility, and safe rollout.
• Monitor production models for drift, degradation, and upstream data quality issues, and own the retraining and refresh cycle.
• Partner with product and engineering to connect model metrics to business outcomes, and document modelling decisions, assumptions, and trade-offs so results can be reproduced and challenged.