Sessions

Hands-On Workshop

This session will dive into the creation of AI Agents using modern frameworks like CrewAI, LangGraph, and more. The goal is to equip all attendees with the knowledge and skills to build systems that leverage multiple tools for complex, real-world tasks. This training will show participants how to design and implement Agentic AI workflows capable of integrating APIs, executing code, generating images, and performing advanced decision-making to accomplish multi-step operations.

Over two hours, you will gain hands-on experience setting up a development environment, designing and lanching agents, crafting chain-of-thought prompts, and creating a custom AI agent framework that automates workflows. The session will also cover optimization and evaluation techniques for agent performance. This session is ideal for intermediate-advanced AI practitioners.

Talk

As AI agents become more capable and complex, ensuring secure, observable, and scalable deployment is no longer optional—it’s foundational. In this session, we’ll explore how open-source tools like Unity Catalog AI (OSS) and MLflow can powerfully support AI agent development with built-in governance and observability.

We’ll cover:

Building agentic systems using LLMs, tool/function calling, with OSS orchestration libraries

Applying access controls, function registration, and sandbox execution using Unity Catalog AI OSS

Adding observability to agents with MLflow Tracing: capturing prompt logic, tool usage, and execution metadata

Addressing production-readiness with strategies for fallback logic, trace debugging, and reproducible agent behavior

By the end of this talk, attendees will understand how to build governed, transparent, and production-ready AI agents—entirely with open-source tools.

Talk

I will present my team's research on rigorously evaluating AI agents. Our findings help explain the gap between the hopes around AI agents and their underwhelming real-world impact so far. I will identify key engineering challenges that need to be solved to deploy agents more successfully. Looking ahead, I will present a framework for thinking about the medium-long term impact of agents on businesses and discuss the steps that can be taken to steer this impact in a beneficial direction.

Talk

Current approaches to agentic systems can be broadly categorized into two groups: Static Chaining, or Prompt & Pray. Both approaches have their own flaws. Static chaining based systems are rigid and can only be applied to narrow solutions. It is also labor-intensive in development and ongoing maintenance. Prompt & Pray is an alternative which gives more decision power to LLM, but it often leads to unreliable results, especially related to tool calling. The lack of accuracy, consistency, robustness plus low maturities of available tools, especially complex context retrieval, are some of the other issues of today’s agentic systems.

Unlock Premium Features with a Subscription

  • Live Tarining:

    Full access to all live workshops and training sessions.

  • 20+ Expert-Led Workshops:

    Dive deep into AI Agents, RAG, and the latest LLMs

  • ODSC Conference Discounts:

    Receive extra discounts to attend ODSC conferences.

Dozens of Free Courses with Premium