Senior Staff Applied ML Engineer - Consumer Agents

About the role

SENIOR STAFF APPLIED MACHINE LEARNING ENGINEER, CONSUMER AGENTS Location: Remote — North America

ABOUT THE ROLE

The Consumer Agent team builds Shopify's consumer-facing agentic shopping experiences — the AI agents inside the Shop app and on merchant storefronts that help millions of buyers search, discover, and buy. As a Senior Staff Applied Machine Learning Engineer, you'll be the person we point at the hardest problems on the surface — distillation to own-GPU inference, agentic memory, whole-page optimization, propensity modeling, or the evaluation rigor that ties them together — and trust to drive it to a shipped, measured outcome. This is a senior individual-contributor seat with deep autonomy: you'll own ambiguous, high-leverage problems end to end, raise the bar for evaluation across the team, and ship weekly on a fast-moving consumer surface. You'll work at the intersection of LLM orchestration, model training and distillation, search and retrieval, and cost/latency engineering, partnering with frontier model labs and ML teams across Shopify.

KEY RESPONSIBILITIES

- Build and improve student models through supervised fine-tuning and knowledge distillation, including the serving, generation-termination, and throughput-economics work that own-GPU inference demands.

Design and raise the bar for ML evaluation across the team — judge design, offline and online metrics, and the judgment to know when a number is real.

Develop agentic memory and personalization signals — buyer and user representations, profile generation — that feed predictive layers.

Work on whole-page optimization: page composition, ranking, and blending for the agent results surface.

Build propensity and predictive models for intent and purchase propensity across agent surfaces.

Move fluidly between distillation, memory, ranking, propensity, and evaluation as priorities shift, without lengthy re-tooling.

Apply rigorous experimentation on a high-traffic surface — proper holdouts, A/B rigor, and counterfactual or off-policy evaluation where online tests aren't enough.

Integrate cleanly with an LLM orchestration stack already in flight — tool calling, structured output, and retrieval-augmented patterns.

Become the technical reference point the team routes its hardest model and evaluation calls to.

Contribute to cross-team ML architecture conversations across Shopify's ML organization and external model-lab partnerships.

Ship weekly, absorbing ambiguity rather than adding coordination overhead.

QUALIFICATIONS

- You've shipped LLM systems to production at real scale — systems with real users, not research prototypes.

Hands-on depth in model training and distillation: supervised fine-tuning and knowledge distillation in practice.

Comfort with the serving and throughput-economics side of inference, not just the training run.

Rigorous evaluation methodology: judge design, offline and online metrics, and the judgment to distinguish a real result from an artifact.

You operate as a senior individual contributor with high autonomy — you take an ambiguous, hard problem and drive it quickly to a measured outcome without hand-holding.

Breadth over a single specialty: you can move across distillation, memory, whole-page optimization, propensity, and evaluation.

Strong communication and collaboration in a fast-paced, cross-functional environment.

Able to work with significant overlap with North America time zones. NICE TO HAVE

E-commerce, marketplace, commerce-search, recsys, or ads background — with instincts for selection bias, position bias, and the gap between offline and online metrics.

Inference and serving internals: vLLM or similar engines, structured and constrained decoding, KV-cache and GPU throughput economics, and latency budgets.

Reinforcement learning experience — RLHF, RLAIF, or preference optimization.

Propensity, ranking, or classical applied ML at scale: gradient boosting, neural ranking, calibration, position-bias correction.

Counterfactual evaluation experience — inverse propensity scoring, doubly-robust estimators, off-policy evaluation.

Experience working alongside or with frontier AI products and agentic systems.