DAIR.AI

OVERVIEW

Advanced AI Agents is a course focused on learning how to design and implement advanced agentic systems.

After completing this course, students will know how to apply best practices and understand the latest design patterns and architectures for building agentic systems.

PREREQUISITES

This course leverages the Python programming language, so it requires basic knowledge of Python.

We primarily use the OpenAI and Gemini models, but you can also use OpenRouter to access other models. For multi-agent orchestration, you will be using Google ADK. We are also leveraging a series of tools like Exa APIs to augment agentic systems with web information.

SYLLABUS

Course Introduction

• Learn to build advanced AI agent systems from scratch using Python.
• Get a complete guide to setting up your development environment with Conda and the necessary API keys.

Building with Agents

• Define what AI agents are and distinguish between simple LLM workflows and complex autonomous agents.
• Understand the core components of an LLM-based agent and how it uses tools to accomplish tasks.

Augmented LLMs

• Discover the Augmented LLM as a foundational block powered by planning, tool use, and memory.
• Build your first single-agent search system from scratch using Python and external APIs.

Prompt Chaining

• Master prompt chaining to break complex tasks into simpler, reliable subtasks for better performance.
• Build a practical report generator that uses a chain of prompts to outline, write, and polish content.

LLM Routing

• Explore LLM Routing to direct user queries to the most appropriate, specialized agent or tool.
• Build a customer service chatbot that classifies and routes requests to the correct handler.

Parallelization

• Learn to use parallelization to significantly improve the speed and efficiency of your agent workflows.
• Build a system that runs content generation and safety checks concurrently to reduce latency.

Multi-Agent Systems

• Transition to advanced multi-agent architectures and learn common design patterns.
• Build a sequential multi-agent system where specialized search, writer, and editor agents collaborate to create effective marketing copies.

Evaluator-Optimizer Agents

• Create self-improving agents using the evaluator-optimizer architecture, a powerful feedback loop pattern.
• Build a system where a critic agent evaluates and provides feedback to a generator agent to iteratively refine its output.

Orchestrator-Worker Agents

• Master the hierarchical orchestrator-worker pattern, where a supervisor agent delegates tasks to multiple worker agents.
• Build a sophisticated multi-agent search system where an orchestrator plans the search and workers execute it in parallel.

Multi-Agent Orchestration

• Learn essential best practices for orchestrating effective multi-agent systems, from starting simple to implementing guardrails.
• Understand the key features of a good agent framework, including flexibility, observability, and tool integration.
• Get started with Google's Agent Development Kit (ADK), the course's framework for agent orchestration.
• Explore ADK's core concepts for managing conversations, including sessions for tracking threads, state for dynamic data, and memory for long-term knowledge.
• Build and test agents using the ADK's built-in web UI for easy interaction and debugging.

AI Agent Evaluation and Debugging

• Learn to systematically evaluate your AI agents by analyzing both the final response and the agent's tool-use trajectory.
• Use Comet's Opik platform to trace, log, and debug complex multi-agent interactions.
• Implement an automated evaluation pipeline to measure performance with metrics like LLM-as-a-Judge.

Deploying AI Agents

• Learn to deploy your AI agents for real-world use, mastering the final step of the development lifecycle.
• Follow a step-by-step guide to deploy your multi-agent system using Google Cloud Run.

--

Here's a subset of companies whose employees have benefited from our courses:

Accrete, Airbnb, Alston & Bird, Amazon, Apple, Arm, Asana, Bank of America, Belong For Me, Biogen, Brilliant, Carebound, CDM, CentralReach, Centric Software, Chime, Coinbase, Digital Green, DoHQ, Elekta, Fidelity Investments, Fivecast, Fulcrum Labs, Google, Guru, Gretel, Harrison Insights, Intel, Intuit, Jina AI, JPMorgan Chase & Co, Khan Academy, KnowBe4, Lawyer.com, LinkedIn, LionSentry, MagmaLabs, MasterClass, Meta, Metopio, Microsoft, Moneta Health, Oracle, OpenAI, Rechat AI, RingCentral, Salesforce, Scale AI, Scribd, Space-O Technologies, Sun Life, TD Bank, TELUS Corporations, Trilogy, TTEC Digital, UniCredit, VaxCalc Labs, Vendr, Walmart, Wolfram Alpha, Zemoso Technologies, Zeplin

Reach out to [email protected] with any questions and for team/student discounts.