Performing near-real-time personalized recommendations at scale with Dapr

Derivco is a global tech company and invention house with over 25 years of experience in the iGaming sector. The organization employs more than 2,000 global experts across Australasia, Southern Africa, Europe, and North America, specializing in technology infrastructure, payment processing, and management platforms for major iGaming brands.

Challenge

Historically, the iGaming industry treated all players uniformly, grouping them by common identifiable characteristics. While this allowed group-level interactions, individual personalization remained impossible. Derivco aimed to transform this using AI/ML capabilities to identify player experience characteristics, deliver personalized recommendations, and enable timely interventions throughout the player lifecycle.

The core challenge was execution speed. Unlike traditional batch-based recommendation systems, this use case demanded near-real-time processing within a narrow window before player disengagement—all while maintaining accuracy and regulatory compliance.

Solution

Derivco built an event-driven, distributed system utilizing:

• Pub/Sub messaging for event distribution
• State Management for data persistence
• Virtual Actors for concurrent processing
• Dapr as the application runtime providing infrastructure abstraction

The architecture processes ML recommendations as CloudEvents triggered through Pub/Sub, with personalized recommendations delivered at the individual player level.

Impact

• System built and deployed to production in under 6 months
• Handles up to 320 million events daily
• Processes over 1,000 events per second through actors
• Achieved up to 80 production deployments per day with zero event loss
• Eliminated boilerplate infrastructure code, accelerating technical debt resolution
• Made previously infeasible legacy system rewrites viable
• Enabled engineers to run complete development environments locally
• Provided clear system visibility through integrated telemetry

The full case study is available on the CNCF website.