Agentic AI system development

>> Available in Danish <<

This course is designed for technical practitioners who want to build agentic AI systems. We start with important AI concepts like context and prompt engineering and programmatic usage of LLMs, then move through ReACT agents, MCP servers and clients, and RAG systems, and finish with security concepts.

Each topic includes a theory introduction, practical code demonstration, and hands-on coding exercises. For agents, agentic concepts, RAG, and MCP, we work with LangChain and LangGraph, but all the knowledge is transferable to other frameworks and libraries.

At the end of the course, you will have a solid understanding of agentic AI systems—their limitations and strengths—and the knowledge to build an agentic system for yourself or your organization.

Participant profile:

This course is aimed at those who have a basic understanding of Python, such as writing simple scripts and being familiar with data structures. More advanced topics, libraries, and projects will be taught in class.

Course content:

• Short theoretical introduction to the transformer architecture and LLMs
• Strong and weak sides of Large Language Models
• Overview of the current AI landscape (frameworks, libraries, models, user interfaces) and terminology
• Programmatic use of Large Language Models

• Improving LLM reliability
• Prompt engineering techniques
• Introduction to context engineering for Agents

• Introduction to LangChain, LangGraph
• LangGraph development model
• Building ReACT agents, tool use
• Multi Agent architectures - overview, strengths, drawbacks, implementation
• Context engineering, Memory, Persistence

• Connecting AI Agents to external tools using MCP
• Developing our own MCP servers

• Implementing an entire RAG pipeline: Loading Documents, Indexing, Storing, Retrieval and Generation
• Building RAG workflows with LangGraph
• Improving RAG performance using advanced techniques:
  • Multi-Query RAG
  • Query Decomposition
  • HyDE (Hypothetical Document Embeddings)
  • Step-back Prompting
  • Routing
  • Multi-Representation Indexing

• Walkthrough of the most common AI Application vulnerabilities
• Techniques to secure an agentic application
• Jailbreaking exercise