This repository contains a high-level overview of a real-time ads recommendation engine built on Google Cloud Platform. The system leverages Vertex AI and BigQuery to deliver personalized ad recommendations and has demonstrated a 35% increase in click-through rates.
- Data ingestion & preprocessing: Pipelines ingest streaming ad events from Pub/Sub, clean and join them with user profiles, and load them into BigQuery for feature engineering.
- Model training: Uses GCP Vertex AI to train and tune ranking models for personalized recommendations, including support for hyperparameter search and managed experiments.
- Deployment: Containerized model serving via Cloud Run exposes a REST API for real-time ad recommendations; integrates with Vertex AI Prediction for scalable deployment.
- Real-time prediction API: Provides a low-latency endpoint for requesting recommendations based on user context and impression history.
- Monitoring & evaluation: Utilizes Cloud Monitoring and custom metrics to track model performance, click-through rates, and detect data drift; supports A/B testing for continuous improvement.
The actual source code and data for this project are proprietary. This repository provides an overview of the system architecture and features. For inquiries, please contact the repository owner.