Agentic workflows combine specialized LLMs, tool usage, and validation techniques to solve complex, real-world tasks. In this session, we will walk through practical design patterns and strategies to build robust multi-agent systems that are scalable, grounded, and capable of self-correction.

We will explore how to structure interactions between agents using routing, sequential chaining, and asynchronous orchestration. Through real-world demos, we’ll show how structured outputs, task guardrails, and grounding with multimodal models (VLMs, audio, OCR) can be combined to ensure reliable performance. This session is hands-on, code-rich, and designed to equip attendees with implementation-ready insights.

The session will provide practical examples and demonstrations of multi-agent systems, including: asynchronously coordinating agents for parallel data processing or project management; sequential agent flows for tasks like document extraction, summarization, and translation; a Router Agent for directing customer support queries; a VLM agent for image analysis and object identification; a Query-to-Dashboard system for generating visualizations from natural language queries; and an audio processing agent that transcribes spoken commands and acts upon them.