boredjnurseman/ml-recommender-system-design

This repository demonstrates a narrow film-recommendation serving workflow: a request uses a deterministic stand-in for cached top-25 candidates, applies regional eligibility, assigns a stable policy cohort, optionally reranks, and returns five films. Policy files supply the versioned weights; disabling reranking preserves the filtered relevance-order fallback. The README also describes offline model production, telemetry, monitoring, and deployment as design context, not implemented subsystems. The graph follows the sampled serving calls and documented policy loading; no deployed API or external store is present.