In this hands-on workshop, participants will explore the practical side of building AI agents using the latest OpenAI technologies. Designed for developers, product professionals, and AI enthusiasts, this 1.5-hour session will walk attendees through the full process of creating, customizing, and deploying an intelligent agent that can reason, interact, and perform tasks autonomously. We’ll start with a quick overview of recent advancements in OpenAI’s ecosystem, focusing on agentic architectures, API integrations, and multimodal workflows – before diving straight into implementation. Together, we’ll build a simple yet powerful AI agent step by step, covering prompt design, context management, and tool usage. By the end of the session, participants will: – Understand how to design and structure an AI agent using OpenAI’s latest features. – Gain hands-on experience integrating real-world data and tools. – Learn best practices for evaluating and refining agent performance. No prior experience with agents is required—just curiosity, basic programming familiarity, and readiness to experiment. Bring your laptop and get ready to build your first AI agent live!

The internet is shifting from websites and apps to AI-native interfaces. In this workshop, Kone.VC, an official OpenAI partner, explores how the new ecosystem of MiniApps and Model Context Protocol (MCP) servers is transforming user interaction, content delivery, and business logic online. We’ll start with an overview of the emerging AI web, where conversational interfaces become the primary gateway for discovery, productivity, and engagement. The second part is a technical deep dive: • Building MCP servers for data integration and control. • Designing and deploying OpenAI MiniApps for personalized user experiences. • Managing content, moderation, and analytics within the ecosystem. Finally, we’ll discuss business and creator benefits — how companies and developers can leverage these tools to launch new products, services, and revenue models in the AI-driven internet. Target audience: Developers, AI engineers, and product builders interested in practical AI integration and the future of digital ecosystems.

In this presentation I will focus on why Human in the Loop is a critical requirement for real world AI adoption. While most discussions today lean toward full automation, the reality inside organizations shows that AI becomes significantly more dependable when a human remains involved at specific decision points. This approach is still not widely recognized or properly framed in the industry. The goal of this session is to bring clarity on how to design AI in a way that actually works in production environments. The presentation will break down Human in the Loop into clear mental models. Where AI can act independently with high confidence and where human judgment is required. I will explain how intervention checkpoints can be structured without slowing down execution, how verification loops and escalation steps should be designed to maintain control without micromanaging the workflow. We will go through examples showing how hybrid AI flows outperform both fully manual operations and fully autonomous pipelines. The audience will understand how this model reduces repetitive review cycles while still keeping accountability and context with humans. To help the audience understand this concept better, I will also walk through a short working build of a Human in the Loop AI flow end to end. This build will make it easier for people to see how these principles translate into an operational system. Outcome for attendees: They will gain a grounded understanding of how to design AI environments that are faster, safer and can scale. They will be able to clearly differentiate when AI should operate on its own and when human oversight becomes necessary. The core learning will be about how Human in the Loop is the central structure that makes AI usable and dependable in real world applications.

Medical documentation continues to be one of the biggest productivity bottlenecks in healthcare. In this hands-on workshop, participants will learn how to design and deploy an agentic AI medical scribe that transforms real or simulated doctor–patient conversations into structured SOAP (Subjective, Objective, Assessment, Plan) notes. Using LangGraph, we’ll build a dynamic, multi-agent system where each node performs a specialized task and communicates through a shared memory graph: Transcription Agent — converts clinical dialogue into text using a speech-to-text model. Information Extraction Agent — identifies key clinical entities such as symptoms, vitals, and diagnoses. Summarization Agent — generates structured SOAP notes and validates internal consistency through a self-reflective loop. Attendees will see how LangGraph enables state-aware orchestration and real-time coordination between these agents, how to integrate retrieval grounding using clinical ontologies (ICD-10, SNOMED), and how to measure performance with F1, ROUGE-L, and readability metrics. By the end, participants will deploy a lightweight, production-ready prototype that demonstrates how agentic workflows can automate time-consuming documentation tasks—while remaining transparent, auditable, and compliant.