Course Outline
Day 1
Anatomy of a Modern AI Agent
Defining autonomous reasoning and acting systems distinct from traditional chatbots
Analysis of reactive, proactive, hybrid, and goal-directed agent paradigms
Essential components: perception, planning, memory, tool integration, and execution
Evaluating design tradeoffs between single-agent and multi-agent architectures for government applications
Agent Frameworks and the Modern Stack
Assessment of LangChain, LlamaIndex, AutoGen, and CrewAI in relation to specific operational requirements
Comparative analysis with established frameworks such as JADE and SPADE
Criteria for selecting appropriate frameworks based on production standards and scalability needs
Mechanisms for tool calling, function invocation, and structured data outputs
Practical exercise: Developing a single Python-based agent with integrated tool capabilities
Multi-Agent System Architectures
Structural models: centralized, decentralized, hybrid, and layered Multi-Agent Systems (MAS)
Communication protocols: FIPA ACL, message-passing methodologies, and contemporary equivalents
Coordination strategies including planning, negotiation, and synchronization mechanisms
Analyzing emergent behaviors and self-organization within agent populations
Decision-Making and Learning in Agents
Application of game theory to cooperative and competitive agent interactions
Implementation of reinforcement learning within multi-agent environments
Mechanisms for transfer learning and knowledge sharing across distributed agents
Strategies for conflict resolution and establishing trust among coordinating agents in sensitive contexts
Day 2
Multi-Modal Foundations for Agents
Unified workflows integrating text, image, speech, and video modalities
Evaluation of leading multi-modal models: GPT-4 Vision, Gemini, Claude, and Whisper
Fusion techniques for synthesizing multiple data inputs within an agent’s reasoning process
Optimizing latency, cost efficiency, and accuracy in multi-modal processing pipelines
Building the Perception Layer
Image processing capabilities: classification, captioning, and object detection
Speech recognition utilizing Whisper Automatic Speech Recognition (ASR) and streaming transcription
Text-to-speech synthesis for natural language interaction interfaces
Integration of perception outputs with Large Language Model (LLM) reasoning and tool selection logic
Hands-On - Building a Multi-Modal Agent in Python
Defining agent objectives, context windows, and available tool inventories
End-to-end integration of GPT-4 Vision and Whisper APIs
Implementation of memory retention, state management, and conversation history control
Safe implementation of tool calls capable of producing verifiable real-world outcomes
Hands-On - Orchestrating a Multi-Agent System
Composing specialized agents using AutoGen or CrewAI frameworks
Establishing roles, responsibilities, and inter-agent communication protocols
Resource allocation and coordination strategies within simulated operational environments
Comprehensive logging of agent reasoning, tool execution, and decision pathways for auditability
Day 3
Threat Surface of Production AI Agents
Differentiating vulnerabilities in agentic AI systems compared to traditional software infrastructure
Identification of attack vectors across data, model, prompt, tool, output, and interface layers
Threat modeling methodologies for agent-based systems with autonomous capabilities
Comparative analysis of AI cybersecurity practices against established information security standards
Adversarial Attacks Hands-On
Execution of adversarial examples and perturbation techniques: FGSM, PGD, and DeepFool
Evaluation of white-box versus black-box attack scenarios
Analysis of model inversion and membership inference vulnerabilities
Mitigation strategies for data poisoning and backdoor injection during the training phase
Defense against prompt injection, jailbreaking, and unauthorized tool misuse in LLM-based systems
Defensive Techniques and Model Hardening
Implementation of adversarial training and data augmentation strategies
Utilization of defensive distillation and other robustness enhancement techniques
Input preprocessing, gradient masking, and regularization methods
Application of differential privacy, noise injection, and management of privacy budgets
Federated learning and secure aggregation protocols for distributed training environments
Hands-On with the Adversarial Robustness Toolbox
Simulating attacks against the multi-modal agent developed on Day 2
Measuring system robustness under perturbation and quantifying performance degradation
Iterative application of defenses with continuous re-evaluation of attack success rates
Stress-testing tool-call pathways and evaluating vulnerability to prompt injection vectors
Day 4
Risk Management Frameworks for AI
Overview of the NIST AI Risk Management Framework: govern, map, measure, manage
Alignment with ISO/IEC 42001 and emerging AI-specific standards
Integration of AI risk metrics into existing enterprise Governance, Risk, and Compliance (GRC) frameworks
Requirements for accountability, auditability, and comprehensive documentation in government systems
Regulatory Compliance for Agentic Systems
Analysis of the EU AI Act: risk categorization, prohibited uses, and obligations for high-risk systems
Implications of GDPR and CCPA regulations on agent data pipelines and processing
Alignment with the U.S. Executive Order on Safe, Secure, and Trustworthy Artificial Intelligence
Sector-specific compliance guidance for finance, healthcare, and public service domains
Management of third-party risk associated with supplier AI tool utilization
Ethics, Bias, and Explainability
Identification and mitigation of bias within agent perception and reasoning processes
The role of explainability and transparency in maintaining security integrity
Fairness considerations, potential for downstream harm, and principles for responsible deployment
Designing inclusive, auditable, and equitable agent behaviors
Production Deployment, Monitoring, and Incident Response
Secure deployment architectures for single and multi-agent systems
Continuous monitoring protocols for model drift, anomalies, and potential abuse
Maintenance of logging, audit trails, and forensic readiness for agent actions
Development of AI security incident response playbooks and recovery procedures
Analysis of case studies involving real-world AI breaches and derived lessons learned
Capstone and Synthesis
Comprehensive review of the multi-modal multi-agent system constructed throughout the course
End-to-end pipeline evaluation: design, development, security hardening, governance, and deployment
Self-assessment of system alignment with NIST AI RMF functions for government use
Analysis of emerging trends in agentic AI and the future landscape of AI security
Summary and Next Steps
Requirements
Testimonials (3)
The trainer is patient and very helpful. He knows the topic well.
CLIFFORD TABARES - Universal Leaf Philippines, Inc.
Course - Agentic AI for Business Automation: Use Cases & Integration
Good mixvof knowledge and practice
Ion Mironescu - Facultatea S.A.I.A.P.M.
Course - Agentic AI for Enterprise Applications
The mix of theory and practice and of high level and low level perspectives
