Operating clinical AI at an annualized volume exceeding 1.2 trillion input tokens requires a robust governance framework and resilient developer infrastructure. Concurrence’s architecture, built on Databricks, demonstrates how scaling complex, safety-critical healthcare agents depends on immutable event sourcing, unified catalog governance, and centralized AI access control.
- Scales to 1.2 trillion yearly tokens with secure AI model governance
- Immutable event histories enable reliable patient state computations
- Simulation-driven testing decouples agent evaluation from live data
Infrastructure signal
Concurrence’s platform runs approximately 11.2 million LLM calls every month, processing over 100 billion input tokens in 30-day periods, scaling to a trillion-token annual rate. This volume growth—around fivefold since late 2025—demands robust data governance and reliable agent infrastructure. The foundation is a data lakehouse architecture built on Databricks, utilizing governed Delta tables and Lakebase for operational state serving.
A key infrastructure innovation is the immutable event sourcing model that treats patient data as append-only events rather than overwriting records. This enables precise provenance tracking and consistent patient state computation. Event streams—including millions of world-model events monthly—flow through Zerobus Ingest into these governed transactional tables, assuring reliability and compliance across evolving workflows.
Developer impact
Developers benefit from centralized AI workload management with Unity Gateway, which integrates secure, governed AI access controls across a fast-growing volume of clinical agent calls. This federation creates a shared foundation for deploying AI models while enforcing compliance policies throughout the development lifecycle.
Importantly, Concurrence has instituted a rigorous simulation and testing pipeline before agents interact with live patients. Simulated patient traffic currently exceeds production traffic by a factor of seven. Immutable patient histories allow teams to replay and test agent responses offline, reducing operational risk and accelerating iteration speed. This shift displaces legacy prompt-logging and reverse-ETL processes in favor of streamlined governance and observability.
What teams should watch
Teams focusing on clinical AI development should prioritize robust event-driven patient state management, ensuring all updates preserve prior context and source details. This approach supports traceability and composability of AI decisions, which is critical for high-stakes healthcare environments. Observability layers that link agent conversation traces, clinical data, and evaluation results in unified data tables are essential for investigating performance and compliance outcomes.
Operationally, consolidation on a governed lakehouse platform enables retiring specialized infrastructure like prompt logs and reverse-ETL, simplifying compliance and reducing cloud cost overhead. However, as workloads grow rapidly, maintaining centralized AI access and strict controls via solutions like Unity Gateway is crucial to uphold security without slowing developer workflows.