Retrieval + ranking pipeline for Avito ad recommendations. Processes 5B events under a 10 GB RAM budget. Co-visitation beats ALS 3.7x on a novelty-filtered target.

Large-scale e-commerce recommendation engine with memory-efficient co-visitation, leakage-safe evaluation, feature engineering, and LightGBM LambdaRank reranking.

A reproducible, resource-aware solution for the Kaggle OTTO multi-objective recommendation competition, featuring co-visitation retrieval, target-aware candidates, LambdaMART ranking, and neural retrieval experiments.