Inventory Intelligence, Enterprise SaaS Architecture, and Application Recovery With Dreeze

Oded Kedem, CTO and co-founder at Dreeze, presented an in-depth look at the “under-the-hood” aspects of their platform, focusing on inventory intelligence, enterprise SaaS architecture, and application recovery. The presentation began by detailing Dreeze’s approach to inventory intelligence, which involves a strategic blend of deterministic code and AI. Data is first scanned and enriched from the cloud using API-based code. Individual resources are then classified, often utilizing smaller AI models to determine characteristics like data type or environment (e.g., production vs. dev), complementing existing tags. Larger AI models are employed for understanding complex applications, identifying software dependencies, and mapping out the full application landscape, while basic dependencies are handled by code. Finally, code aggregates all learned information to calculate impact scores (importance of a resource/application for backup) and risk scores (recovery readiness), which help customers prioritize testing and resource investment, with Dreeze currently leveraging Anthropic models for its AI capabilities.

The discussion then shifted to Dreeze’s robust Enterprise SaaS architecture, designed specifically for enterprise needs rather than adapting consumer SaaS models. This architecture is built around a dynamic “cell” system, where each cell is a multi-tenant environment serving a specific task. There are authentication cells for centralized login, control cells that handle the bulk of the work and can be dedicated private cells for large enterprises (ensuring GDPR compliance and housing AI models via AWS Bedrock), and data cells that are co-located within the customer’s region to minimize data transfer and security concerns. A core principle is that customer data remains within their own cloud accounts, and Dreeze connects via role assumption, making it easy to integrate and disconnect. This complex yet flexible architecture provides resilience, compliance, minimizes costs, and centralizes authentication without burdening the end-user with underlying complexities.

Finally, Kedem elaborated on Dreeze’s approach to application recovery and the internal challenges of building such a system. All elements necessary for recovery—data, metadata, and recovery plans—are stored in an immutable, self-sufficient vault within the customer’s environment, in open, encrypted formats owned and accessible by the customer. The platform automates the creation of a recovery plan from the application graph, allowing customers to inject custom steps before execution. Building Dreeze presented significant challenges due to the vast number of cloud objects and resources, each requiring specific handling. Kedem emphasized that such an autonomous and comprehensive solution would be impossible without AI to cope with the immense scope. To ensure reliability, Dreeze employs an extensive testing system that utilizes real cloud accounts, with AI agents continuously running tests, identifying bugs, and facilitating fixes around the clock. The key takeaways underscore the importance of using AI judiciously, ensuring sensitive data never leaves customer premises, and the critical role of managing and testing the recovery graph for reliable recovery.