Compound AI systems: unboxing a reference architecture
Modern AI systems are increasingly complex, requiring seamless integration of diverse components, robust governance, and scalable infrastructure.
Compound AI Systems, as defined by Berkeley's AI Research Lab, tackle AI tasks using multiple interacting components, processes, and tools.
In this session, we will introduce a structured approach to designing, deploying, and managing enterprise-grade AI solutions using Union’s reference architecture. Attendees will learn how to unify modular components into cohesive systems while addressing critical challenges like collaboration, security, and real-time observability.
The session will begin with an overview of the reference architecture, highlighting its core pillars:
- Actors: long-running "warm" containers for extremely fast executions.
- Artifacts: Managing inputs, outputs, and reusable assets (datasets, model versions, pipelines) with traceability.
- Serving: Deploying scalable inference endpoints and APIs while optimizing latency, cost, and reliability.
- RBAC: Implementing granular role-based access control to secure sensitive data and models.
- Live Debugging: Monitoring systems in production, diagnosing failures, and iterating without downtime.
A live demo will bring these concepts to life, showcasing how Union’s tools enable teams to build, govern, and refine compound AI systems end-to-end. Whether you’re an ML engineer, architect, or platform lead, this session will equip you with actionable strategies to simplify complexity, enforce governance, and accelerate time-to-value in AI initiatives.
Who should attend
AI/ML practitioners, platform engineers, and technical leaders seeking to operationalize robust, collaborative, and auditable AI systems.
About the Speaker
Pablo is a Solutions Architect at Union AI, focusing on ML/AI pipelines. Before joining Union he spent several years at Databricks and DataRobot, immersed in data science and engineering. Before that, he spent about 20 years in the software industry, working mostly at software vendor companies in everything from virtualization to application performance. He is originally from Argentina but has been in Los Angeles for a long time.
About Union.ai
Union is an AI platform that simplifies ML infrastructure so you can develop, deploy, and innovate faster.
Write your code in Python, collaborate across departments, and enjoy full reproducibility and auditability. Union lets you focus on what matters.
💬 Join our AI and MLOps Slack Community: slack.flyte.org
⭐ Check out Flyte on GitHub: github.com/flyteorg/flyte
🤝 Get access to $30 in GPU credits and a hosted Flyte platform signup.union.ai