Loading…
Loading…
Loading…
Loading…
Go from the agentic loop to governed, production multi-agent systems: tool use, planning, memory, MCP, agentic RAG, multi-agent orchestration, safety, sandboxing, observability, and enterprise deployment.
Haithem
Instructor
Paid items use bank transfer. After checkout, full payment steps appear in My Orders (login required). We activate your access after we verify your receipt.
Secure checkout · Card or bank transfer
This course includes:
Building an AI agent that works in a demo is easy; building one that operates safely, predictably, and cost-effectively in production is a genuinely different engineering discipline. This course takes you from the fundamental observe-reason-act loop through tool use, planning, memory, and reasoning frameworks (ReAct, Tree of Thought, self-critique), into multi-agent orchestration, browser and computer-use agents, agentic coding assistants, and the safety, sandboxing, observability, and governance practices that make autonomous systems trustworthy.
You'll build agents with structured tool use, agentic retrieval, human-in-the-loop confirmation gates, injection-resistant safety controls, sandboxed execution, distributed tracing, and multi-agent orchestration — capped by two capstones: a single autonomous multi-tool agent, and a governed multi-agent platform defended live against a reviewer including a real safety-incident drill.
ML/AI engineers, backend engineers moving into agentic systems, and product engineers building on top of LLM APIs. Comfortable with Python and familiar with LLM prompting basics (covered in this catalog's NLP and Generative AI courses); no prior agent-framework experience required — this course builds every concept from first principles through enterprise production practice.
-67%FeaturedGo from first principles to running production NLP systems: tokenization, embeddings, transformers, fine-tuning, RAG, prompt engineering, alignment, evaluation, safety, and enterprise-grade serving.
AI Agents & Multi-Agent Systems Engineering: Zero to Hero
49.99 TND
What makes a system an 'agent' rather than a single LLM call
The autonomy spectrum and choosing the right level for a task
Function calling: structured tool invocation
Handling tool failures and building a tool execution layer
Task decomposition: breaking a goal into sub-tasks
Detecting and recovering from plan failure
Short-term working memory and context window management
Long-term and episodic memory across sessions
The ReAct pattern: interleaving reasoning and acting
Self-critique, reflection, and iterative refinement
Why agent-tool interoperability needed a standard protocol
Building and consuming an MCP server
Agentic RAG: retrieval as a tool, not a fixed pipeline stage
Grounding multi-step agent decisions in retrieved evidence
Tree of Thought: exploring multiple reasoning paths
Cost-aware deliberation: when to use expensive reasoning strategies
Why multiple agents instead of one more capable agent
Sequential, hierarchical, and broadcast communication patterns
Graph-based orchestration: expressing agent workflows explicitly
Comparing orchestration approaches: graph-based, conversational, and role-based
Designing well-scoped, differentiated agent roles
Manager-worker delegation in practice
Deciding where human confirmation checkpoints belong
Designing effective human review interfaces for agent actions
Task success rate and trajectory evaluation
Cost efficiency, robustness testing, and continuous agent evaluation
Why prompt injection is more dangerous for agents than single-turn systems
Tool misuse, resource exhaustion, and abuse detection
Least-privilege permission scoping for agent tool access
Sandboxed execution environments for agent actions
Why agent cost and latency are variable, and how to budget for it
Optimization techniques: model routing, parallelization, and caching for agents
Checkpointing and resumability for long-running agents
Progress monitoring and intervention for extended autonomous operation
Repository-scale reasoning for coding agents
Safety and evaluation for code-executing agents
Browser agent architecture: perceiving and acting on web pages
Safety and reliability for autonomous web browsing agents
GUI automation beyond the browser: full computer-use agents
Evaluating and operating computer-use agents reliably
Distributed tracing for multi-step agent execution
Dashboards and alerting for agent-specific health signals
Graceful degradation and intelligent retry strategies
Self-healing patterns: automated diagnosis and correction
Building deterministic simulation environments for agent testing
Unit testing agent components in isolation
Structured multi-agent debate for error correction
Resolving conflicting conclusions between cooperating agents
Versioning and staged rollout for agentic systems
Scaling agent infrastructure: concurrency, queuing, and multi-tenancy
Audit trails for autonomous agent actions
Compliance considerations specific to autonomous agentic systems
Agents versus traditional RPA: when agentic automation actually helps
Integrating agents into existing enterprise systems
Enterprise agent case studies: coding, support, and research automation
AI agent engineering interview preparation
Capstone I: Multi-Tool Autonomous Agent
Capstone II: Multi-Agent System with Human Oversight & Live Defense
No reviews yet. Be the first to review!
Enroll in this course to join the discussion.
No comments yet. Start the discussion!
FeaturedA 180-hour beginner-to-expert AI engineering program covering software, data, ML, deep learning, vision, NLP, LLMs, RAG, agents, MCP, MLOps, cloud, security, architecture, and AI SaaS.