Lead pCTR modeling research for Instacart's ads ranking and retrieval systems

Instacart is hiring a research-leaning senior ML engineer to advance pCTR/conversion prediction models, tackle training-data bias (selection, position, optimizer's curse), and help build next-gen foundation-model and generative retrieval systems (TIGER, Semantic ID) for ads. This interview goes deep on causal inference, debiasing techniques, and multi-task/sequence model architecture — not infra or full-stack engineering.

Step into this interview

Free · a live voice mock calibrated to this exact role

Practice this interview

What this interview tests

pCTR/conversion prediction modeling and calibration
Causal inference & training-data bias mitigation (selection, position, optimizer's curse)
Multi-task, multi-domain model architecture (MoE, LoRA, transformers)
Generative retrieval & sequence modeling (TIGER, Semantic ID)
Formulating ambiguous ML problems into scoped research directions

Common question themes

How would you detect and correct position bias in ads click-through training data?

Explain the difference between Platt scaling and isotonic regression for model calibration — when would you pick one over the other?

Design a Multi-Domain Multi-Task architecture that shares a backbone across ad surfaces while allowing domain-specific fine-tuning.

How does a generative retrieval system like TIGER differ from embedding-based nearest-neighbor retrieval, and what new failure modes does it introduce?

Walk through how you'd formulate a vague product observation (e.g., ads feel overcalibrated for new advertisers) into a rigorous research question with a measurable evaluation plan.

View the original posting