(covering all Universities from the North to the south of Tunisia )
Building Agentic AI Workflows with xAI - From Code to Action
This repository contains the full hands-on material used in the Google Vertex AI Agent Workshop Series, delivered during the Tunisia AI Universities Tour 2026 .
It demonstrates how to build agentic AI systems step by step, starting from a local AI agent and progressing toward cloud-scale agentic workflows on Google Vertex AI, with a strong focus on explainability (xAI) and responsible AI.
Tunisia AI Tour’26 is a nationwide educational journey bringing hands-on Agentic AI workshops to universities across Tunisia.
From data to deployment, participants do not just learn AI, they build real intelligent agents using industry tools such as Google Vertex AI.
Beyond technology, the tour is rooted in impact, with 50 percent of workshop proceeds donated to people in need through the Tunisian Red Crescent.
Education, access, and responsibility moving together across the country. 🇹🇳✨
Google vertex-AI agent-workshop/
│
├── local-ai-agent/ # Demo 1: Local agent (Ollama + Python)
│ ├── agent.py
│ ├── llm.py
│ ├── main.py
│
├── agent-demo (VScode) demo 2/ # Demo 2: Agentic AI with prediction + actions
│
├── Vertex AI Agent Builder/ # Vertex AI managed agent approach
│
├── Presentation/ # Slides used during the workshop
├── Ressources/ # Reading material and references
│
├── GOOGLE VERTEX AI DEMO TAKEWAY.pdf
├── ressources links.pdf
└── README.md
By following this repository, you will learn how to:
- Understand Agentic AI vs Generative AI
- Build a multi-step autonomous AI agent
- Separate LLM reasoning from agent control
- Implement planning, memory, autonomy, and stopping logic
- Add explainability (xAI) to agent decisions
- Scale agents to Vertex AI pipelines and Agent Builder
This repository is designed for:
- Students
- Developers
- Data scientists
- AI engineers
- Educators running hands-on workshops
The workshop follows a progressive architecture, starting locally and scaling to Google Vertex AI.
flowchart TD
U[User Goal or Prompt]
A[AI Agent Control Layer
Planning
Memory
Autonomy
Stopping Logic]
LLM[LLM Reasoning Engine
Ollama or Vertex LLM]
T[Tools and Actions
APIs
Models
Databases
External Systems]
XAI[Explainable AI Layer
SHAP
Reason Logs
Auditability]
O[Final Output or Action]
U --> A
A --> LLM
A --> T
LLM --> A
T --> A
A --> XAI
XAI --> O
- The LLM is not the agent
- The agent owns planning, memory, and decisions
- Explainability is built in by design
- The same structure works locally and in production
- This mirrors real enterprise AI systems
This demo runs 100 percent locally using Ollama and Python. No API keys. No cloud required.
[Steps for installation, code, and execution remain exactly as in the previous version]
An AI assistant that:
- Collects data from APIs
- Runs predictions using a trained model
- Explains decisions using xAI techniques
- Acts autonomously
- Logs every step for traceability
This section shows how the same agent architecture maps directly to Google Cloud Vertex AI.
flowchart TD
U[User or Application]
AG[Vertex AI Agent Builder
Agent Logic
Orchestration]
M[Vertex AI Models
Gemini
Custom Models]
P[Vertex AI Pipelines
Workflow Automation]
FS[Feature Store
Data Sources]
XAI[Vertex Explainable AI
SHAP
Attribution
Monitoring]
ACT[Actions
APIs
Cloud Functions
External Systems]
LOG[Logging
Monitoring
Governance]
U --> AG
AG --> M
AG --> P
AG --> ACT
M --> XAI
P --> LOG
ACT --> LOG
XAI --> LOG
-
Agent Builder Manages agent logic, autonomy, and orchestration
-
Vertex AI Models LLMs and predictive models used for reasoning and inference
-
Vertex Pipelines End to end workflow automation and execution
-
Explainable AI Transparency, trust, and regulatory readiness
-
Governance and Monitoring Logging, auditing, and production safety
| Local Workshop Component | Vertex AI Equivalent |
|---|---|
| Python Agent | Agent Builder |
| Ollama LLM | Vertex LLMs |
| Local Memory | Feature Store |
| Python Logic | Pipelines |
| Logs | Cloud Logging |
| SHAP | Vertex Explainable AI |
- Chatbots respond
- Agents decide
- xAI explains
- Production systems audit everything
- https://docs.cloud.google.com/agent-builder/overview
- https://cloud.google.com/vertex-ai/docs
- https://codelabs.developers.google.com/building-ai-agents-vertexai
- https://cloud.google.com/discover/what-are-ai-agents
- https://bbycroft.net/llm
- https://n8n.io
- Senior AI Engineer , Netflix USA
- AI Expert Consultant , Tegus USA
- PhD Candidate in Explainable AI (xAI) , Saint MAry's University Canada
- International AI Speaker ( Linkedin Top AI Voice )
- CEO and Head of AI R&D at Man.AI Global
- AI Expert Reviewer & Author , Packt Publishing UK
- Social Media : https://taplink.cc/manaimortadha
- LinkedIn Profile : https://www.linkedin.com/in/mannai-mortadha/
- GitHub Portfolio : https://github.com/MortadhaMannai
- Leetcode Profile : https://leetcode.com/u/mannaimortadha898/
- Meduim Blog : https://www.google.com/url?q=https://manaimortadha.medium.com/
- Sessionize Profile: https://www.google.com/url?q=https://sessionize.com/Mortadha_Mannai
- Email : mannaimortadha898@gmail.com
Tunisia AI Universities Tour 2026 Building Agentic AI Workflows with xAI on Google Vertex AI - From Code to Action