Key Takeaways From Week 3 of the AI Builders Summit — AI Agents

We wrapped up week 2 of our first-ever AI Builders Summit! With hundreds of people tuning in virtually from all around the world, our world-class instructors showed how to make the most out of your AI agents. Here’s a recap of each session from this week, and if you feel like you’re missing out, then you can still sign up for next week’s sessions on building AI and also get these sessions on-demand.

You can also read the recaps of the previous weeks here and here.

AI Agents — A Practical Implementation

Valentina Alto, Technical Architect, AI & App at Microsoft

Valentina Alto explored the rapid evolution of generative AI over the past two years, highlighting the transition from simple text generation to the rise of AI agents. She provided a detailed breakdown of AI agent architecture, emphasizing components such as memory, knowledge bases, and tool integration. A key focus was on the paradigm shift from traditional conversational AI to agentic applications capable of orchestrating complex tasks autonomously. The session included a hands-on demonstration of building an AI agent from scratch, using blockchain for orchestration. She also introduced retrieval-augmented generation (RAG) and vector search, explaining how embeddings enhance contextual understanding in AI agents. The talk concluded with best practices for designing AI-driven applications, balancing autonomy with responsible AI principles.

Building Agentic RAG with LlamaIndex Workflows

Laurie Voss, VP of Developer Relations at LlamaIndex

Laurie Voss provided a deep dive into LlamaIndex, a framework for connecting structured data to LLMs, enabling efficient retrieval-augmented generation (RAG) and multi-agent orchestration. He guided participants through implementing AI agents using LlamaIndex, covering agent workflows, state management, and human-in-the-loop interactions. The session featured a step-by-step workshop on building a multi-agent system, where agents collaborated to research, write, and review reports. Voss also introduced workflow-based orchestration, enabling developers to construct flexible, modular AI applications with branching logic, parallel execution, and self-reflective agents. Emphasizing scalability, he showcased LlamaIndex’s ability to integrate with vector databases and various LLM providers, streamlining enterprise AI adoption.

Modern AI Agents from A-Z: Building Agentic AI to Perform Complex Tasks

Sinan Ozdemir, AI & LLM Expert | Author | Founder + CTO at LoopGenius

Sinan Ozdemir presented a comprehensive overview of agentic AI, outlining the foundational concepts, design principles, and key frameworks for building autonomous AI systems. He dissected the four core elements of AI agents: task definition, tool usage, rules and constraints, and structured prompts. The session covered different paradigms for agentic behavior, including planning-based approaches, reflection mechanisms, and dynamic tool invocation. Ozdemir emphasized that while existing frameworks (e.g., CrewAI, LangChain) provide out-of-the-box agent functionality, developers can construct their own lightweight agentic systems by focusing on modularity and flexibility. His talk concluded with best practices for deploying AI agents in real-world applications, ensuring robustness, adaptability, and ethical AI considerations.

Using World Models to Build AI Agents for Optimal Decision-Making

Dr. Andre Franca, CTO of Ergodic

Andre explored the concept of world models as a framework for AI agents to make optimal decisions in uncertain environments. He emphasized the need for AI agents to map the world before taking actions, representing states as contextual knowledge that evolves over time. The session introduced a simulation-based approach where AI agents iteratively explore possible actions, evaluate their outcomes, and refine decision-making processes. Franca demonstrated the use of LLMs as world models by building a simple reinforcement learning agent that adapts based on historical data, using a stepwise state-action transformation. He also discussed the limits of reinforcement learning in agentic AI, advocating for hybrid models that combine symbolic reasoning with data-driven learning to enhance real-world applicability.

LLM Engineering Masterclass: Select and Apply LLMs Using RAG, Fine-tuning, and Agentic AI

Edward Donner, Co-founder and CTO of Nebula.io

Ed Donner provided a deep dive into LLM engineering, covering the selection, optimization, and deployment of large language models (LLMs) for practical AI applications. He introduced multi-agent AI workflows, where different LLMs collaborate in decision-making. The session explored RAG (Retrieval-Augmented Generation), fine-tuning with QLoRA, and agentic AI frameworks to enhance model performance. Donner showcased a multi-step LLM pipeline using DeepSeek, integrating 23 different LLM calls per decision to create a highly autonomous AI agent. A hands-on workshop followed, where participants built an e-commerce AI assistant capable of dynamically selecting the best LLM for various NLP tasks. He emphasized cost-performance tradeoffs, evaluation techniques, and real-world deployment considerations.

Building and Evaluating an Agentic RAG Application with LangGraph

Greg Loughnane, Co-Founder & CEO at AI Makerspace

Chris Alexiuk, Head of LLMs at AI Makerspace | Founding Machine Learning Engineer at Ox

Chris and Greg co-presented a session on the practical implementation of AI agents in business applications. Ozdemir focused on structured agent workflows, breaking down AI tasks into decision trees with modular components. He discussed autonomous agents using rule-based and machine learning-driven approaches, highlighting the importance of self-reflection mechanisms in AI workflows. Donner built on this by demonstrating multi-agent collaboration, where LLMs handle different roles (e.g., reasoning, verification, retrieval) to optimize decision-making. The session concluded with a live coding demo, where an AI agent was deployed in a customer segmentation use case, leveraging causal inference, reinforcement learning, and real-time data adaptation for personalized marketing strategies.

Conclusion

On February 5th and 6th, we’ll be wrapping up the AI Builders Summit with a series of workshops and quick demos showcasing how to use many of the tools featured throughout the event.

• AI Q&A + Live-Speed-Build with Matt Shumer
• From Idea to Implementation: How to Self-Host an AI Agent
• Cloning NotebookLM with Open Weights Models
• Built a Data Analyst AI Agent from Scratch
• Building AI Agent Workflows: Automating Research Papers to Podcasts
• Build your own, better LLM-powered Slackbot and Save Thousands
• Real-World Workflows for Solving Everyday Problems with AI
• Run DeepSeek-R1
• Red Hat OpenShift AI — Predictive and Generative AI Demo
• Advancing GraphRAG: Multimodal Integration with Associative Intelligence

You can register now to catch next week’s sessions and even see everything from the past three weeks on demand! If you’re looking for even more hands-on AI training, then you can register for ODSC East this May 13th-15th and get access to the AI Builders Summit included!