Agentic AI

Introduces the core concepts of Agentic AI, the course structure, prerequisites, and learning environment.

Learn what AI Agents are and how they work. Understand the critical role prompting plays in guiding them to reason, plan, and act to achieve goals.

Explains the theory of using roles or personas to control the tone, style, and expertise of an LLM's output.

Provides hands-on practice in iteratively developing a role-based prompt to create a believable historical figure persona.

Explains the conceptual frameworks for Chain-of-Thought (CoT) for guided reasoning and ReAct (Reason+Act) for enabling agents to plan and take actions.

Provides hands-on practice implementing both CoT and ReAct prompts to solve a retail analytics problem.

Explains the theory of systematically refining prompt instructions by modifying components like Role, Task, Context, Examples, and Output Format.

Provides hands-on practice iteratively refining a prompt to transform a generic recipe analyzer into a precise dietary consultant that produces structured JSON.

Explains the conceptual framework for building multi-step AI workflows by linking the output of one prompt to the input of the next, and the importance of validation.

Provides hands-on practice implementing a three-stage prompt chain with Pydantic-based gate checks to automate an insurance claim triage process.

Explains the conceptual framework for building self-improving systems where an agent uses feedback from its own actions to iteratively refine its output.

Provides hands-on practice building an automated feedback loop where an AI generates Python code, has it tested against a unit test suite, and uses the test results as feedback to debug itself.

Course review

Introduces the foundational concepts of AI agents and agentic workflows, setting the stage for the course. It covers prerequisites, the course environment, and how to use the necessary API keys.

Explores what defines a modern AI agent, its core components (Persona, Knowledge, Tools, Interaction), and the different types of agents based on their LLM interaction model.

Design and visualize agentic workflows. Learn common agent types as building blocks for creating visual workflow diagrams.

Design and visualize agentic workflows. Learn common agent types as building blocks for creating visual workflow diagrams.

Covers the practical aspects of translating agentic workflow models into Python code. Students learn to structure agent logic, define agent classes, and orchestrate their interactions.

Introduces the Prompt Chaining pattern for breaking down complex tasks into a sequence of smaller, dependent steps. It covers strategies for task decomposition, validation, and context management.

Provides hands-on experience in implementing the Prompt Chaining pattern. Students build a multi-agent chain to solve a problem where information is passed sequentially.

Teaches the Routing pattern, which involves classifying incoming tasks and directing them to the most appropriate specialized agent or processing path.

Students implement a routing system where a router agent uses an LLM to classify a query and then dispatches it to the correct specialist agent, which may involve orchestrating sub-tasks.

Introduces the Parallelization pattern for executing multiple agent tasks concurrently. It covers strategies for task decomposition (sharding, aspect-based) and result aggregation.

Students implement a parallel workflow using Python's threading module, where multiple specialist agents analyze a document concurrently, and a synthesizer agent combines their findings.

Focuses on the Evaluator-Optimizer pattern, an iterative process of generation, critique, and refinement to improve output quality. It emphasizes clear evaluation criteria and actionable feedback.

Students build a two-agent system (a creator and a critic) that works in a loop. The creator generates a solution, and the critic provides feedback until the solution meets all constraints.

Introduces the advanced Orchestrator-Workers pattern, where a central agent dynamically plans, delegates, and synthesizes the work of multiple specialized worker agents.

Students implement a market analysis report generator where an Orchestrator agent creates a plan, assigns tasks to news, competitor, and trend analysis workers, and then synthesizes their findings.

Course review.

Get to know your course instructors, set up OpenAI resources, and get an overview of the course.

Extend AI agents beyond text with tool integrations, enabling reliable real-time actions and data access.

Develop AI agents in Python using tools with OpenAI SDK. Interact through language models, build functionality-enhancing tools, and test via tool-augmented exercises.

Discover structured outputs in AI: transform responses into actionable JSON for integration. Utilize schemas, parsers, and function calls to enhance reliability and automation in workflows.

Master structured outputs with Pydantic and OpenAI SDK for LLMs. Learn parsing, type validation, and create validated AI agent responses in JSON format.

Explore agent state management with state machines. Learn how agents track user input, instructions, and tool use for complex workflows, ensuring adaptability and reliability.

Master Python state machines: set up environment, define schemas, manage transitions, and run workflows. Explore advanced routing and loops for dynamic workflows.

Explore short-term memory in AI agents, enhancing coherence via state, ephemeral, and ephemeral memory strategies for efficient context retention in active sessions.

Learn to implement short-term memory in Python for coherent AI interactions via a ChatBot with personas, enabling session continuity and dynamic responses.

Explore using external APIs for real-time data, dynamic actions, and authenticating agents. Discover MCP, a protocol standardizing AI’s tool interoperability and safety.

Explore using OpenAI and Python to integrate external APIs, make GET/POST/PUT requests, manage API keys, and create agents for real-time data interactions.

Equip agents to search web for real-time, unstructured info. Ground responses in evidence using APIs, handle noise, and avoid hallucination for credible answers.

Build a web search agent using Python, Tavily API, to integrate real-time web data, parse results, and enhance language models' effectiveness.

Equip agents to access and modify structured data by using SQL for interaction and vector databases for semantic tasks, ensuring seamless integration with private systems.

Convert natural language to SQL using SQLAlchemy, SQLite, and text2SQL Agent to interact with databases efficiently through real-world examples and practical applications.

Discover Agentic RAG: Enhance RAG by enabling reflection, query reformulation, and intelligent adaption for nuanced answers. Master retrieval, reasoning, and retry loops.

Explore agentic RAG in Python using ChromaDB, integrating AI with retrieval-augmented generation for intelligent document retrieval and processing with OpenAI embeddings.

Explore long-term agent memory: understand semantic, episodic, and procedural memories. Learn storage strategies and best practices for personalized, coherent interactions.

Implement long-term memory in Python agents using vector databases for enhanced user interaction, session persistence, and personalized responses.

Agent Evaluation guides assessing an agent’s task completion, quality, tool use, and system metrics using response, step, or trajectory strategies to ensure reliable and efficient operations.

Evaluate Python-based agents by setting environments, creating tools, designing test cases, and using diverse evaluation methods to enhance performance and design.

Congratulations on completing the course!

Learn the core concepts of multi-agent systems and their real-world parallels.

Explain the core components of multi-agent systems and how to design their high-level architecture.

Explain the core components of multi-agent systems and how to design their high-level architecture.

Develop a multi-agent system by coding the designed architecture and connecting agents with well-defined interfaces.

Develop a multi-agent system by coding the designed architecture and connecting agents with well-defined interfaces.

Apply orchestration techniques to coordinate multiple agent actions and achieve complex workflows.

Apply orchestration patterns (sequential parallel conditional) to build a multi-agent system that handles complex workflows with multiple steps and decision points.

Configure routing mechanisms to manage data flow among agents in multi-agent systems.

Implement a routing agent that can intelligently direct user requests to specialized agents based on the content and urgency of the request.

Evaluate methods for tracking and updating agent state across multi-turn interactions.

Evaluate methods for tracking and updating agent state across multi-turn interactions.

Develop a coordinated multi-agent system that synchronizes states for coherent task execution.

Implement a multi-agent system that manages concurrent access to shared resources using state coordination techniques to detect and resolve conflicts.

Extend RAG to multiple cooperating agents, each specialized in certain retrieval tasks.

Build a multi-agent RAG system with specialized retrieval agents and a synthesis agent to combine information from multiple sources and make a complex judgment.

Course review