Write clean Python
Git, tests, APIs and databases: the foundations of an engineer, not just a ChatGPT user.
First cohort · limited seats
Become an AI Engineer
from Git to agents in production
A guided 18-week path through Python, APIs, language models, RAG and AI agents. Every step ends with a real project you build and publish, all the way to an app you put online.
- Starts
- November 2026
- Length
- 18 weeks
- Seats
- 20 seats
- Format
- Online · live
By the end
What you’ll actually be able to do
Pick the right model
Fast, reasoning or open-weight: you know which one to use, for which task, at what cost.
Build RAG and agents
Assistants that read your documents, call tools and chain several steps.
Ship a real app
UI, Docker, deployment, monitoring and security: your project has a URL and users.
The roadmap
9 steps, from Git to your Demo Day
Each step builds on the one before and ends with a project published on your GitHub.
01Week 1 Git and your setup
Why now · Everything you build goes through Git: it’s your history, your portfolio and how you work with a team.
You learn- Terminal and VS Code
- Git: commit, branch, merge
- GitHub: pull requests, issues
- Markdown and READMEs
You buildYour first public repository with a polished README, the start of your portfolio.
02Weeks 2–4 Python for engineers
Why now · Python is the language of AI: model SDKs, data tools and agent frameworks are all written in Python.
You learn- Types, lists, dicts, functions
- Modules, packages and environments (uv)
- Files, JSON and errors
- HTTP requests and async
- Testing with pytest
You buildA command-line tool that queries a public API and saves the results.
03Weeks 5–6 APIs and data
Why now · An AI app is an app first: an API, a database and clean data.
You learn- HTTP and REST
- FastAPI and Pydantic
- SQL and PostgreSQL
- pandas essentials
You buildA REST API with its database, documented and tested.
04Week 7 Understanding LLMs
Why now · Before building on a model you need to know what’s under the hood, and how to pick the right model for each task.
You learn- Tokens, context and temperature
- Fast models vs reasoning models
- API models vs open-weight models run locally
- Cost, latency and limits
- Embeddings
You buildAn “arena”: the same tasks sent to several models, compared on quality, speed and cost.
05Week 8 Prompting and structured output
Why now · In production a good prompt isn’t enough: you need reliable JSON, tools called correctly and measured results.
You learn- System prompts and few-shot examples
- Structured output (JSON Schema)
- Function / tool calling
- Streaming
- First evaluations
You buildAn extractor that turns CVs or invoices into validated JSON.
06Weeks 9–10 RAG: AI on your documents
Why now · A model doesn’t know your documents: RAG gives it the right information at the right time, with sources.
You learn- Chunking and embeddings
- Vector database (pgvector)
- Hybrid search and reranking
- Citing sources
- Evaluating a RAG system
You buildAn assistant that answers from your course material in Arabic and French, citing its sources.
07Weeks 11–13 AI agents
Why now · An agent plans, uses tools and takes action: this is where AI becomes a real product.
You learn- The agent loop: plan → tool → observation
- Tools and MCP (Model Context Protocol)
- Memory and state
- Workflows vs autonomous agents
- Guardrails and human approval
You buildAn agent that uses several tools (search, database, calculator), plus your own MCP server.
08Weeks 14–15 Shipping to production
Why now · A demo on your laptop isn’t a product: you need to deploy, monitor and keep costs under control.
You learn- A UI with React / Next.js
- Docker and deployment
- Logs, traces and evals in CI
- Costs, caching and rate limits
- Security: prompt injection and secrets
You buildYour app online, with a real URL to share.
09Weeks 16–18 Capstone and Demo Day
Why now · A complete product presented to a jury is what convinces a recruiter, and it can become your PFE.
You learn- Scoping and architecture
- Code review with a mentor
- Demo and pitch
- Portfolio, CV and GitHub
You buildA complete AI product, presented at Demo Day.
The models
The right model for each task
There is no single “best” model. You learn to compare the families on the same tasks, then choose on quality, speed and cost.
Fast models
They answer right away. Great for extracting, classifying, summarising and chatting.
Speed
Reasoning
Cost
- Best for
- Extraction, classification, chat
- Example
- Sorting 10,000 customer messages
Reasoning models
They think before answering: slower and pricier, but much better at maths, code and multi-step plans.
Speed
Reasoning
Cost
- Best for
- Planning, code, analysis
- Example
- Planning an agent’s steps
Open-weight models
Llama, Qwen, Mistral… you run them yourself: private data and predictable cost, but you need the hardware.
Speed
Reasoning
Cost
- Best for
- Sensitive data, large volumes
- Example
- Documents that never leave your server
Embedding models
They don’t generate text: they turn meaning into vectors for search and RAG.
Speed
Reasoning
Cost
- Best for
- Search, RAG, recommendations
- Example
- Finding the right paragraph
During the cohort you test every family on the same tasks, so you choose with numbers, not hype.
Capstone projects
Projects that look like real work
Pick your capstone from these ideas, or bring your own. Each one can become your PFE.
Customer support agent
Answers a Tunisian business’s customers in derja and French, checks orders and hands over to a human.
- RAG
- Tools
- WhatsApp API
PFE assistant
Helps structure the report, checks sources and prepares the jury’s questions.
- RAG
- JSON
- Next.js
CV ↔ job matching
Analyses CVs and job offers and suggests the best profiles, with a justification.
- Embeddings
- Reranking
- FastAPI
Code review bot
Reviews GitHub pull requests and comments on bugs and security issues.
- GitHub API
- Reasoning
- CI
Voice assistant
You speak, it understands and answers out loud: speech-to-text → LLM → text-to-speech.
- STT
- TTS
- Streaming
Data analyst
Ask a question in plain language; the agent writes the SQL, runs it and plots the chart.
- Text-to-SQL
- pandas
- Agents
The format
How the cohort works
2 live sessions a week
A lesson and a live-coding session in the evening, with live Q&A.
A project review every week
A mentor reviews your code and tells you what to improve.
A private community
A group to help each other, share projects and move forward together.
Everything is recorded
Replays, slides and exercises stay available.
FAQ
Before you book
Another question? Message us on WhatsApp, we reply ourselves.
I’ve never coded. Is this for me?
Yes. The roadmap starts with Git and Python, and every step has a project within your reach. What matters most is consistency: about 8 hours a week.
How much does it cost?
The price of the first cohort is shared with the people who book. Booking is free and commits you to nothing.
What computer do I need?
Any recent laptop with 8 GB of RAM. We mostly use models through APIs, so no graphics card is needed.
Can the capstone become my PFE?
Yes, it’s designed for that. And if you want to go further, our PFE mentoring takes over all the way to your defense.
What language are the sessions in?
We explain in Tunisian Arabic and French, and keep technical terms in English, as in the industry.
Booking
Book your spot in the first cohort
Booking is free and commits you to nothing: we’ll contact you to check your level and give you the exact dates and price.
Only 20 seats
- 01You send your booking
- 02We reply ourselves on WhatsApp
- 03Your spot is confirmed before the start