DSM — Retail Recommender Platform
Component DSM of a retailer's recommendation stack showing why an online learning system cannot be fully sequenced: serving, metrics, drift, retraining and the model registry close a loop that absorbs 10 of the 14 components into a single iteration block, leaving only event collection, identity resolution, the feature store and experiment assignment cleanly upstream.
Make it your own.
title "Retail recommender — component coupling"
mode component
convention ir-fad
components: Event collector, Identity resolution, Feature store, Catalogue embeddings
components: Candidate generator, Ranking model, Business rules layer, Guardrail filter
components: Serving API, Experiment assignment, Metrics warehouse, Retraining job
components: Model registry, Drift monitor
Identity resolution <- Event collector(3)
Feature store <- Event collector(3), Identity resolution(3)
Catalogue embeddings <- Feature store(2), Model registry(2)
Candidate generator <- Catalogue embeddings(3), Feature store(2)
Ranking model <- Feature store(3), Candidate generator(3), Model registry(3)
Business rules layer <- Ranking model(2), Catalogue embeddings
Guardrail filter <- Business rules layer(2), Catalogue embeddings(2)
Serving API <- Ranking model(3), Guardrail filter(3), Candidate generator(2), Experiment assignment(2)
Experiment assignment <- Identity resolution(2)
Metrics warehouse <- Event collector(2), Serving API(2), Experiment assignment(2)
Retraining job <- Metrics warehouse(3), Feature store(3), Drift monitor(2)
Model registry <- Retraining job(3), Drift monitor
Drift monitor <- Metrics warehouse(2), Feature store(2)