AI Agents and Automation

At the end of the course, you will earn a Certificate of Completion from Spirit in Projects.

Do you want to learn about the next evolutionary stage of AI – autonomous agents that can independently plan tasks, make decisions, and automate complex workflows? If you want to understand how AI agents work and how you can develop them yourself with modern frameworks like LangChain, CrewAI, or AutoGPT, then look forward to this training. From the theoretical foundations of agent architectures to practical implementation with current OpenSource tools, you will learn how to develop intelligent automation solutions for your company. In extensive hands-on exercises in Google Colab, you will build your own AI agents and multi-agent systems.

Advising

Objectives

  • Understand fundamentals of AI agents: Definition, architecture, autonomy
  • Learn about different agent types and their application areas
  • Develop LLM-based agents with ReAct paradigm
  • Master current OpenSource frameworks (LangChain, CrewAI, AutoGPT)
  • Implement Tool Use and Function Calling for agents
  • Orchestrate multi-agent systems for complex tasks
  • Develop practical automation solutions with AI agents
  • Understand safety, evaluation, and deployment strategies for agents

Target groups:

AI Expert, Software Developer, ML Engineer, System Architect, Software Architect, DevOps Engineer, Automation Specialist and anyone who wants to engage with AI agents and intelligent automation.

Show dates and times

Syllabus

1. Fundamentals of AI Agents

  • What are AI agents? Definition and delineation
  • From simple tools to autonomous systems
  • Agent vs. Model vs. Copilot
  • Autonomy and decision-making
  • Perception, action, goal-oriented behavior
  • Agent-environment interaction
  • Historical development: From rule-based to LLM-based agents
  • Current trends: Growing importance of AI agents in enterprise applications

2. Agent Architectures

  • Reactive agents: Stimulus-response
  • Deliberative agents: Planning and reasoning
  • Hybrid architectures
  • BDI model: Beliefs, Desires, Intentions
  • Layered architectures
  • Subsumption architecture
  • Comparison and application scenarios

3. Planning Algorithms for Agents

  • Goal-based planning
  • Utility-based planning
  • State space search
  • Forward vs. backward planning
  • Hierarchical Task Networks (HTN)
  • Planning algorithms in practice

4. Reinforcement Learning for Agents

  • Fundamentals of Reinforcement Learning
  • Markov Decision Processes (MDP)
  • Q-Learning and Deep Q-Networks (DQN)
  • Policy gradient methods
  • Multi-armed bandits
  • RL for autonomous agents

5. LLM-based Agents

  • Revolution through Large Language Models
  • LLMs as reasoning engines
  • ReAct paradigm: Reasoning + Acting
  • Observation, Thought, Action loop
  • Chain-of-Thought for agents
  • Tree-of-Thought for complex decisions
  • Self-reflection and self-correction

6. Tool Use and Function Calling

  • Concept of Tool Use in LLMs
  • Function Calling: OpenAI, Anthropic, Google approaches
  • Tool description and schemas
  • Multi-tool orchestration
  • Connecting external APIs: Web search, calculator, database
  • Error handling and retry strategies
  • Practical exercise: Agent with multiple tools (Google Colab)

7. Memory and Context Management

  • Short-term vs. long-term memory
  • Conversation history management
  • Memory types: Buffer, summary, entity, knowledge graph
  • Vector databases as long-term memory
  • Context window optimization
  • Practical exercise: Agent with memory system

8. LangChain for Agents

  • Overview of LangChain framework
  • Agents in LangChain: Zero-shot, conversational, ReAct
  • Tools and toolkits
  • Chains vs. agents
  • AgentExecutor and agent types
  • Custom tool development
  • LangSmith for agent debugging
  • Practical exercise: ReAct agent with LangChain (Google Colab)

9. LlamaIndex for Data Agents

  • LlamaIndex fundamentals
  • Data agents for RAG workflows
  • Query engines and data connectors
  • Multi-document agents
  • Integration with vector databases
  • Practical exercise: RAG agent with LlamaIndex

10. Autonomous Agents: AutoGPT and BabyAGI

  • Concept of autonomous task execution
  • AutoGPT: Architecture and functionality
  • BabyAGI: Task-driven autonomous agent
  • Task decomposition and prioritization
  • Iterative task planning and self-critique
  • Limitations and challenges
  • Practical exercise: Autonomous research agent (Google Colab)

11. Multi-Agent Systems

  • Why multi-agent systems?
  • Agent communication and coordination
  • Role-based agent design
  • Collaborative vs. competitive agents
  • Consensus and negotiation
  • Task distribution and load balancing

12. CrewAI for Multi-Agent Orchestration

  • Overview of CrewAI framework
  • Crews, agents, tasks, tools
  • Role-based agent definition
  • Process types: Sequential, hierarchical, consensual
  • Agent delegation and collaboration
  • Output handling and result aggregation
  • Practical exercise: Multi-agent system with CrewAI (Google Colab)

13. Cost Optimization

  • Minimize token usage in agents
  • Caching strategies
  • Model selection: High-end models vs. low-end vs. open source
  • Routing: Simple tasks to cheaper models
  • Batching and parallelization
  • Cost monitoring and budget alerts
Spirit in Projects