Senior Machine Learning Engineer, Search & Recommendations

Overview

The Search & Personalization ML team is Instacart’s engine for state-of-the-art multi-task, multi-objective ranking—unifying search, discovery, recommendation, 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.

About the Job

• Architect the ranking backbone that unifies query understanding, personalization, multi-objective ranking, ads, and merchandising into a single adaptive platform.
• Design and build a search autosuggest system optimized for personalization and value-based relevance.
• 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.