Overview
Recommendation systems fail in predictable ways: they optimize for clicks when the business needs purchases, they perform well on active users and fail on new ones, and they are evaluated offline on metrics that don't correlate with the online business outcome. A system with 0.85 NDCG that drives no incremental revenue is not a good recommendation system — it is a well-evaluated bad one.
Effective recommendation system design requires three decisions beyond algorithm selection: what to optimize (the metric must match the business objective), how to handle cold start (new users and new items have no interaction history), and how to evaluate offline in a way that predicts online performance. Most implementations get at least one of these wrong.
The Recommendation System Design Prompt generates a complete recommendation architecture: algorithm selection by data type and business objective, cold start strategy, hybrid system design, offline evaluation framework, and an A/B test specification that measures the business impact of recommendations.
What you get: - Algorithm selection matrix by data type and scale - Cold start strategy for new users and new items - Hybrid system architecture combining multiple signals - Offline evaluation with business-metric-aligned metrics - A/B test specification for online impact measurement
Built for: ML engineers and data scientists building recommendation systems for e-commerce, content, SaaS, and marketplace platforms.