Talki Academy
🤝

CrewAI: Multi-Agent Orchestration in Production

An intensive technical training for developers and AI practitioners who want to move beyond single-task agents and build autonomous agent teams capable of solving complex, multi-step problems. You will compare CrewAI, LangGraph, and AutoGen, define structured roles and tasks, and ship a production multi-agent pipeline that automates a real business process.

Duration
2 days
Level
Advanced
Price
9.99 EUR/month (all courses included)
Max group
12 participants

What you will learn

+Understand the architectural differences between CrewAI, LangGraph, and AutoGen
+Define agents with precise roles, backstories, and goals that drive LLM behavior
+Compose sequential and parallel tasks with dependency management
+Integrate external tools (web search, code execution, APIs, databases) into a crew
+Implement human-in-the-loop checkpoints on critical decisions
+Deploy a crew to production with monitoring, retry logic, and cost observability

Course program

Module 1: CrewAI vs. LangGraph vs. AutoGen: Choosing Your Framework

2h30
  • Multi-agent system anatomy: orchestrator, workers, shared memory, message bus
  • CrewAI v0.80+: role-based philosophy, YAML config, sequential vs. hierarchical process
  • LangGraph: explicit state graph, nodes, conditional edges, checkpointing
  • AutoGen 0.4: multi-agent conversation, GroupChat, AssistantAgent vs. UserProxyAgent
  • Comparative benchmark: autonomy, control, debuggability, cost, adoption
  • Workshop: implement the same pipeline in CrewAI and LangGraph — compare code and traces

Module 2: Defining Agents, Roles, and Tasks in CrewAI

3h
  • Agent: role, goal, backstory — how each parameter shapes LLM behavior
  • Task: description, expected_output, context (task dependencies), output_file
  • Crew: sequential process (chained tasks) vs. hierarchical (delegating manager)
  • Agent memory: short-term (in-crew), long-term (vector), entity (named entities), contextual
  • Built-in CrewAI tools: SerperDevTool, FileReadTool, ScrapeWebsiteTool, CodeInterpreterTool
  • Workshop: build a Monitor/Analyst/Writer crew that auto-produces a weekly AI news digest

Module 3: Custom Tools and External Integrations

2h30
  • Build a custom CrewAI tool with the @tool decorator and BaseTool
  • Pydantic-typed inputs/outputs for automatic validation
  • REST API wrapper: integrate Notion, Airtable, HubSpot, or Slack as agent tools
  • Sandboxed code execution with CodeInterpreterTool and Docker isolation
  • Database access from agents: SQLite (local), PostgreSQL (prod), DynamoDB (serverless)
  • Tool error handling: ToolException, automatic retry, escalation to human-in-the-loop

Module 4: Complex Pipelines and Human-in-the-Loop

3h
  • CrewAI Flows: orchestrate multiple crews with kickoff(), kickoff_for_each(), router()
  • Shared state between crews: Pydantic BaseModel as flow state, inter-step persistence
  • Hierarchical crew: manager_llm, automatic delegation, sub-task evaluation
  • Human-in-the-loop: human_input=True on critical tasks, CLI interactive vs. HTTP webhook
  • Full real-world case: lead qualification pipeline (Discovery → Scoring → Outreach → human validation → CRM update)
  • Timeout management and circuit breakers for long-running pipelines

Module 5: Production: Deployment, Monitoring, and Cost Optimization

2h
  • Langfuse integration: trace every thought, tool call, and LLM call in a crew automatically
  • Key metrics: cost per run, tokens per agent, tool error rate, P95 latency
  • Cost optimization: assign claude-haiku-4-5 to routing agents, claude-sonnet-4-6 to writers
  • Retry logic: max_iter, max_rpm (rate limiting), 429/500 handling with exponential backoff
  • Async deployment: AWS Lambda (cold start < 3s) + SQS for long-running crews (> 30s)
  • Alerting: cost spike > 2× baseline, error rate > 5%, timeout detection

Module 6: Capstone Project: Automate a Full Business Process

1h
  • Specify and deliver an autonomous crew on a real business case
  • Group code review: architecture, security, cost, robustness
  • 5-minute live demo of the crew with execution metrics
  • Trainer feedback and personalized post-training roadmap

Ready to get started?

9.99 EUR/month — All courses included, cancel anytime

Request a quoteView all courses