Building Agentic Systems

Content

This training is aimed at developers and software architects who want to develop agentic AI systems for data processing and business process automation. The course is framework-independent and teaches fundamentals, concepts, and decision criteria through practical examples.

The course consists of the following topics and may be extended or adapted based on the audience. For in-house courses there is a selection of technologies and scenarios which can be chosen to better fit the audience.

– Fundamentals of Generative AI and Large Language Models:

• Overview of the current LLM landscape and model types
• Tokens, context windows, and model properties
• Prompt engineering techniques and best practices
• API integration: OpenAI API, Azure OpenAI, local models
• Hands-on code examples and building a simple LLM harness

– The Model Context Protocol (MCP):

• MCP architecture and concepts
• Implementing MCP servers and clients
• Tool definitions and resource provisioning via MCP
• Integrating MCP into existing applications
• Hands-on: MCP server for a business process

– Retrieval Augmented Generation (RAG):

• RAG architecture and design patterns
• Embedding models and vector databases
• Chunking strategies and indexing
• Hybrid search and reranking
• RAG pipeline optimization and evaluation
• Hands-on: RAG system for document processing

– Self-Hosting AI Models:

• Overview: When self-hosting, when cloud APIs?
• Ollama and local inference for development and production
• ONNX Runtime and model formats
• Quantization and performance optimization
• Deployment strategies for self-hosted models

– Model Evaluation and Selection:

• Evaluation criteria: Quality, latency, cost, data privacy
• Benchmarking methods for LLMs
• Systematic model comparison for specific use cases
• Evaluation frameworks and metrics
• Hands-on: Model comparison for a concrete use case

– Fine-Tuning Assessment and Strategy:

• When fine-tuning, when RAG, when prompt engineering?
• Decision matrix for the right approach
• Fine-tuning concepts: LoRA, QLoRA, and parameter-efficient fine-tuning
• Data preparation and training pipelines
• Cost-benefit analysis of fine-tuning

– AI Agent Fundamentals and Patterns:

• Agent concepts: Perception, reasoning, action
• Agent patterns: ReAct, planning, reflection
• Tool calling and function calling
• Orchestration and control flow
• Hands-on: Building a simple agent harness and agent for data processing

– Multi-Agent Systems and Workflows:

• Multi-agent architectures and communication patterns
• Orchestration vs. choreography
• Agent coordination and task distribution
• Human-in-the-loop integration
• Hands-on: Multi-agent workflow for a business process

– Data Processing and Business Processes with AI:

• Document processing and extraction
• Structured data analysis with LLMs
• Workflow automation with AI agents
• Integration into existing business processes
• Error handling and robustness in AI pipelines

– Architecture and Production:

• Architecture patterns for agentic AI systems
• Security: Input validation, content filtering, guardrails
• Logging, monitoring, and observability
• Testing strategies for AI systems
• Scaling and performance considerations

This course combines theoretical knowledge with intensive practical application and prepares you to make informed decisions and develop production-ready agentic AI systems. Topics are accompanied by code examples and hands-on exercises.

The actual course content may differ from the above depending on the trainer, delivery, duration and the composition of participants.