Intelligent recommendation engine combining preference learning, habit detection, and context awareness to deliver personalized user suggestions powered by LLM endpoints. User commands, screen time data, usage patterns, and external APIs (Calendar, Email, Analytics) flow through a WAF-protected API Gateway into three parallel processing streams that feed an event bus. The Recommendation Engine queries a Vector DB for embeddings, invokes an LLM Endpoint with safety guardrails, and caches responses while persisting user preferences and behavior patterns across dedicated databases. This architecture enables real-time personalized responses, proactive suggestions, and intent-driven adjustments with full observability and monitoring. Fork and customize this diagram on Diagrams.so to adapt the processing pipeline, storage layers, or LLM integration for your use case. The modular design supports scaling individual components—preference learning, habit detection, and context awareness—independently based on traffic and inference latency requirements.