Together with our client from the German automotive industry, we at ZEISS built a fully serverless, event-driven SaaS platform on AWS to deliver standardised production KPIs across a global manufacturing landscape of ~120 factories. From the outset, the system was designed to scale per tenant while providing near real-time reporting, with aggregations available within seconds of machine events.

Apache Iceberg forms the core data layer, bridging streaming ingest and analytical access. Filtered machine signals (~5k/min/tenant) are enriched with critical context from external systems such as shifts, asset management or fault and measures management (~400/min/tenant). A key challenge was that this contextual data could change retroactively, requiring historical aggregations to be corrected. Iceberg’s ACID semantics enabled efficient updates and recomputation while keeping a consistent business view.

We will share the architectural trade-offs and lessons learned from building this system end to end, including what worked and what did not. Delivered jointly by a solution architect and a requirements engineer/business analyst, the talk connects business constraints with practical Iceberg design decisions.