At TRM Labs, we process petabyte-scale blockchain data across 40+ networks to deliver real-time intelligence to financial institutions, governments, and enterprises. As our need for low-latency, high-throughput analytics grew, our legacy serverless data warehouse architecture began to show its limitations.

In this talk, we'll share how we reimagined our data platform—migrating to a modern lakehouse architecture built on Apache Iceberg and powered by StarRocks to support sub-second, user-facing analytics. We'll cover the core design principles behind the platform, the architectural evolution of our StarRocks deployment, and the operational insights we've gathered over a year of production usage.

---

**What You'll Learn**

**1. Replatforming to a Lakehouse Architecture**

- How we transitioned from BigQuery and Postgres to an open, modular lakehouse stack.
- Apache Iceberg's role as the foundational storage layer.
- Why we selected StarRocks as our low-latency analytical engine.

**2. Powering Real-Time Customer Analytics**

- How we use StarRocks to serve customer-facing dashboards and APIs.
- Handling concurrency and diverse query patterns.
- Integrating StarRocks with our upstream event-driven pipelines.

**3. Production Lessons from One Year in the Wild**

- Best Practices: Managing clusters, enabling cross-environment reads, and accelerating developer workflows.
- Performance & Cost Optimization:
- Dual-layered Kubernetes architecture for separating ingestion and serving workloads.
- Advanced data cache tuning for Iceberg-backed StarRocks queries.
- Dynamic cluster scaling and compute-storage decoupling.
- Future Focus: Scaling to support 100+ TPS APIs and hundreds of TBs with StarRocks + Iceberg.
- Tips & Tricks: Versioning, release automation, and extending StarRocks through internal development.