Senior Machine Learning Engineer - Search & Recommendations Ranking

Overview

This is a general posting for multiple Senior Machine Learning roles open across our 4-sided marketplace. You’ll get the chance to learn about the problems the different ML teams solve as you go through the process. Towards the end of your process, we’ll do a team-matching exercise to determine which of the open roles/teams you’ll join. You can find a blurb on each team at the bottom of this page.

The Search & Recommendations ML team is Instacart’s engine for multi-task, multi-objective ranking—unifying search, discovery, ads, and merchandising into a single value-aware platform. Partnering with world-class engineers, scientists, and PMs, we build the ranking backbone that powers every pixel of the shopping journey, optimizing not just for clicks, but for incremental GTV, basket lift, and retention over the long run.

What We’re Building

• Foundational Ranking Backbone Models: Multi-task/multi-objective models (shared encoders + task heads) that jointly learn relevance, conversion, margin contribution, churn risk, and ad quality, enabling consistent decisions across search and recommendations.
• Value-Aware Optimization: Uplift and long-horizon value models that steer decisions toward incrementality and LTV, with calibrated constraints on quality, diversity, fairness, and spend pacing—plus guardrails for safe exploration.
• LLM-Enhanced Retrieval & Features: Using LLMs to enrich query and item semantics for long-tail recall, generate features for cold-starts, and feed the ranker with reasoning-rich context, while remaining the source of truth for final ordering.

Our commitment to AI innovation is reflected in our recent publications and research contributions to the field.

About the Job

• Architect the ranking backbone that unifies query understanding, personalization, multi-objective ranking, ads, and merchandising into a single adaptive platform.
• Design long-horizon objective functions (e.g., incrementality, LTV, habit formation) and build uplift/causal value models that move beyond short-term engagement.
• Develop production-grade Multi-Task Learning (e.g., shared encoders, MMOE/PLE task heads) to jointly learn relevance, propensity, margin, and churn risk—ensuring calibration, constraints, and explainability.
• Own the inference layer: goal-aware re-rankers, diversity and quality constraints, safe exploration, and millisecond-class latency optimization.
• Advance evaluation practices: online experiments, long-horizon cohort metrics, counterfactual evaluations, and attribution pipelines for tracking incremental GTV and retention.
• Partner across ads, infrastructure, product, and design teams to translate business goals into ranking policies and measurable ROI.
• Mentor ML engineers to build expertise in ranking, causal inference, and scalable serving systems.

About You
Minimum Qualifications

• 5+ years applying ML at scale (3+ years in technical leadership), with a proven track record improving ranking or recommendation systems in production.
• Demonstrated success in applying multi-objective or constrained optimization to balance relevance, revenue, margin, and user experience; experience with online testing and attribution beyond CTR.
• Strong coding (Python) and data fluency (SQL/Pandas), with expertise in classic ML techniques (e.g., XGBoost) and deep learning frameworks (TensorFlow/PyTorch).
• Excellent analytical skills and strong cross-functional communication abilities.

Preferred Qualifications

• Expertise in multi-task learning architectures (e.g., MMOE/PLE, shared encoders), calibration, counterfactual evaluation, uplift/causal modeling, and/or contextual bandits for exploration.
• Experience building low-latency ranking services, including feature stores, caching, vector + lexical retrieval, re-ranking, and A/B testing infrastructure, with expertise in constraint-aware inference.
• Hands-on experience with LLMs as feature/recall enhancers (e.g., embeddings, adapter tuning) while maintaining clarity on when the ranker should arbitrate.

#LI-Remote