My role
Machine learning scientist — retrieval, ranking, and evaluation.
Overview
Developed a candidate-generation and ranking stack for on-site recommendations. Trained two-tower embeddings for retrieval, followed by a learning-to-rank model that blends relevance with business objectives. Served embeddings from an approximate nearest-neighbour index for sub-100ms retrieval, and evaluated offline with NDCG before confirming lift in a live A/B test.
Role & impact
Impact
Lifted click-through rate by 27% and added 9% to items-per-order in a controlled experiment.