Target Audience

• Intermediate to advanced Python developers
• Professionals who want to design autonomous AI agents
• Developers who want to automate complex workflows using LangChain, LangGraph and multi-agent systems

Learning Objectives

• Explain the architecture of an AI agent (perception–decision–action loop) and differentiate it from a chatbot
• Implement processing chains and conversational memory using LangChain
• Design a complete RAG system (embeddings, vector database, semantic search)
• Create custom tools and ReAct agents capable of executing business actions
• Orchestrate complex workflows using LangGraph (graphs, branching, human-in-the-loop)
• Coordinate specialized multi-agent systems using CrewAI
• Deploy a complete automated workflow addressing a real business need in production

Prerequisites

• Intermediate Python knowledge (OOP, modules, REST APIs)
• Basic knowledge of LLM APIs (prompt engineering, API calls)
• Completion of the Python & AI (API) training or equivalent level
• A computer with Python 3.x and internet access

Teaching Methods

• Combination of theory (25%) and practice (75%)
• Interactive lectures and live demonstrations of agent architectures
• Individual and team practical exercises using LangChain, LangGraph, CrewAI and real APIs
• Active learning: iterative design, collaborative debugging, group architecture reviews
• Digital course materials and exercise files (permanent access)

Target Certification: RS6962 – Programming and Automating Tasks with Python (TOSA) – CPF eligible

Detailed Program

DAY 1 — Introduction to AI Agents and LangChain

• AI agent concept: difference between a chatbot and an autonomous agent
• Perception–decision–action loop: agent architecture
• Introduction to LangChain: philosophy, components, installation
• Language models in LangChain: configuration and parameters
• Prompt templates: creation, variables, composition
• Chains: operation pipelines, LLMChain, SequentialChain
• Conversational memory: memory types and persistence

■ Practical exercises:
→ Lab 1: Configure LangChain with different LLMs
→ Lab 2: Build chains for multi-step text processing
→ Lab 3: Chatbot with persistent conversational memory

DAY 2 — RAG and Vector Databases

• RAG (Retrieval-Augmented Generation): principles and architecture
• Embeddings: vector representation of text and models
• Document chunking strategies and overlapping
• Vector databases: ChromaDB, Pinecone, Weaviate — comparison
• Document indexing: PDF, Word, web pages
• Semantic search: cosine similarity, top-k retrieval
• Integrating RAG in LangChain: RetrievalQA

■ Practical exercises:
→ Lab 1: Build a ChromaDB vector database from documents
→ Lab 2: Semantic search system on technical documentation
→ Lab 3: RAG chatbot capable of answering from a knowledge base

DAY 3 — Function Calling and Tools

• Function calling: principles, function definition, JSON schema
• Tools in LangChain: creation, documentation, binding
• Built-in tools: web search, calculator, Wikipedia
• Custom tools for accessing business APIs
• ReAct agents: reasoning and action loop
• Error handling for tools and fallback strategies
• Security: validating calls and limiting permissions

■ Practical exercises:
→ Lab 1: Agent using calculation and web search tools
→ Lab 2: Create custom tools (weather, database)
→ Lab 3: Assistant capable of executing business actions (CRM, tickets)

DAY 4 — LangGraph and Complex Workflows

• LangGraph: execution graphs, nodes and edges
• States and transitions: managing data flow
• Conditional branching: dynamic routing
• Loops and iterations in graphs
• State persistence: checkpoints and execution recovery
• Human-in-the-loop: human validation within workflows
• Graph debugging and visualization

■ Practical exercises:
→ Lab 1: Document processing workflow with validation
→ Lab 2: Customer support agent with conditional escalation
→ Lab 3: Content generation pipeline with iterative review

DAY 5 — Multi-Agent Systems and Final Project

• Multi-agent architecture: coordination and communication
• CrewAI: defining roles, tasks and teams
• Agent specialization: research, analysis, writing
• Orchestration: sequential vs parallel execution, dependencies
• Conflict management and consensus between agents
• Monitoring and observability of multi-agent systems

■ Practical exercises:
→ Lab 1: CrewAI team for news monitoring and summarization
→ Lab 2: Multi-agent report writing system
→ Lab 3 (Final project): Complete automated workflow solving a real business problem

Assessment

• Initial assessment test at the beginning of the training
• Formative assessments: corrected practical exercises, architecture reviews, collaborative debugging, intermediate quizzes
• Final 40-question multiple-choice test (Day 5) — passing score: 60%
• Certificate of completion issued