Real-time event-driven systems are easy to sketch on a whiteboard. Idempotent, fault-tolerant, cost-efficient, and maintainable by design, not by accident. Getting them right in production is a different story entirely.

This session shares hard-won lessons from building a production serverless pipeline on AWS that processes high-volume CDC events from Kafka, evaluates business rules against accumulated state, and persists results to a relational database, all without managing servers.

We’ll cover the decisions that actually matter at this layer: how to filter noise before it reaches your business logic, how to design idempotency that survives restarts, retries, and clock drift, when to reach for DynamoDB versus a relational database in the same pipeline, how Kafka and SQS failure semantics differ fundamentally, and why that changes how you write error handling code.

Generative AI is rapidly transforming how modern cloud-native products are designed, developed, and improved. In this session, I will share real-world experience from building and evolving two production-grade platforms at Cloudwalker, a Personalizer Recommender System and a CRM Customer Journey platform, and demonstrate how we successfully integrated Generative AI using AWS services.

The talk will focus on practical challenges we faced while scaling personalization and marketing automation systems, including incomplete metadata, time-consuming content creation, and maintaining safe and high-quality user communication. I will explain how we leveraged Amazon Bedrock and serverless AWS architecture to solve these challenges through automated metadata enrichment, AI-assisted multi-channel content generation, and real-time content safety validation.