← InfraDraft

Data & Analytics

Design a Recommendation Engine (Batch vs. Streaming pipelines)

A hybrid recommendation platform combining offline batch model training with a low-latency online serving path for real-time personalization.

Open and Simulate this Architecture in InfraDraft

Core Architectural Components

Batch Feature Pipeline

Offline ETL producing training features from historical data.

Model Training Pipeline

Collaborative filtering or embedding models trained on a schedule.

Feature Store

Keeps online and offline features consistent between training and serving.

Candidate Generation Service

Narrows the full catalog down to a manageable candidate set per user.

Real-Time Ranking/Scoring Service

Scores and orders candidates within the request's latency budget.

Streaming Event Ingestion

Feeds fresh signals — clicks, purchases — into near-real-time updates.

A/B Testing & Experimentation Framework

Measures whether a new model actually improves outcomes.