Agentic AI Systems: Designing and Deploying Autonomous Workflows
Scheduled sessions
The next evolution of AI: Move beyond passive chatbots and standard RAG pipelines. Learn to build Agentic AI systems capable of reasoning, planning, and executing complex tasks autonomously.
Master Tool Calling & ReAct: Teach Large Language Models how to interact with the outside world—querying databases, calling internal APIs, browsing the web, and executing code.
Multi-Agent Orchestration: Get hands-on with cutting-edge frameworks like LangGraph and AutoGen to build teams of specialized AI agents that collaborate, debate, and resolve complex workflows under a supervisor model.
Safety & Production: ~70% hands-on labs focusing on building reliable agents, implementing Human-in-the-Loop (HITL) safeguards, and preventing infinite loops or rogue execution.
Who it’s for: Senior Software Engineers, AI Researchers, and Architects looking to deploy the next generation of autonomous enterprise applications.
Skills You Will Learn
Curriculum
Foundations of Agentic AI & Tool Calling
- The shift from Generation to Action: Defining an AI Agent
- The ReAct framework (Reasoning + Acting): How LLMs plan their steps
- Function/Tool Calling deeply explained: Schemas, execution, and returning results to the LLM
- Lab: Building a single agent from scratch that queries a SQL database and a live weather API
Stateful Agents & Memory Management
- Handling long-running tasks: Short-term vs. long-term memory
- Semantic memory: Integrating vector databases for agent context
- Handling execution errors: Self-correction and fallback prompting
- Lab: Creating a data-analysis agent that writes, executes, and fixes its own Python code
Multi-Agent Orchestration (LangGraph & AutoGen)
- Why Multi-Agent? Routing complexity to specialized narrow agents
- Introduction to LangGraph: Treating agent workflows as cyclical graphs (StateGraph)
- Microsoft AutoGen: Conversational patterns between autonomous agents
- Supervisor patterns vs. Hierarchical teams vs. Peer-to-peer chat
- Lab: Building a multi-agent software development team (Coder, Reviewer, Tester) with LangGraph
Production, Safety, and Human-in-the-Loop
- Safety first: Sandboxing code execution and restricting API scopes
- Human-in-the-Loop (HITL): Pausing agent graphs for manual approval before critical actions
- Observability: Tracing agent thoughts and tool calls (LangSmith / Phoenix)
- Lab: Deploying an agentic workflow as a streaming API with HITL interruption points
Optional modules
Optional — Advanced Reasoning Patterns
- Tree of Thoughts (ToT) and Monte Carlo Tree Search for LLMs
- Plan-and-Solve architectures for long-horizon planning
- Fine-tuning open-source models specifically for tool calling
Course Day Structure
- Part 1: 09:00–10:30
- Break: 10:30–10:45
- Part 2: 10:45–12:15
- Lunch break: 12:15–13:15
- Part 3: 13:15–15:15
- Break: 15:15–15:30
- Part 4: 15:30–17:30