Build real systems with Claude Code

Four days in Oran. You wire Claude Code into your own company's systems and leave with something running in production — and the working knowledge to keep extending it.

OranIn person

4 daysTwo weekends

Small groupSelected

65,000DZD

To be announcedDates

Request a place See the program

The same skills run these

What you learn here is not locked to one tool.

Claude

ChatGPT

Cursor

Gemini

Copilot

Windsurf

Replit

Codeium

Ollama

LangChain

n8n

Plus Codex, Antigravity, Hermes and OpenClaw.

SkillsMCPSubagentsTeams WorkflowsHooksContextUltracode

This is not an AI introduction, and it is not no-code.

These are the tools I use every day on live client work and my own products. Four days of building, not four days of slides. If you want an overview of what AI can do, this is the wrong room.

Raslem Charmati

AI systems developer · Algeria's first AI automation agency

I work with Algerian businesses to implement AI securely and build them custom systems. What I teach here is the same work I do every day.

Secure AI implementation
Custom systems
Training programmes
Workshops
Business audits

Fifteen pieces that lock together

None of these is useful alone. The workshop is about how they compose — which is the only reason anyone gets a system out the other end instead of a demo.

Context engineering

How context actually works and why quality collapses when it fills. The CLAUDE.md hierarchy across global, project and directory. Compacting versus clearing, and running to a context budget on purpose.

Token economics

Where tokens really go, how prompt caching changes the arithmetic, what a working session costs in dinars, and the specific habits that cut it without costing you output.

Skills

Writing your own, testing them, and making them fire at the right moment. Progressive disclosure and why a well-built skill raises accuracy while lowering cost.

MCP

Wiring Claude to your database, your API and your internal tools — then building your own MCP server from nothing.

Subagents

Context isolation and delegation. Where they genuinely save time and money, and where they quietly burn both.

Teams & parallel agents

Several agents working at once across git worktrees, and coordinating them so they don't overwrite each other.

Workflows

Deterministic orchestration — fan-out, pipelines, verification stages. When a workflow beats a single agent, and when it's overhead.

Efforts & ultracode

Reasoning levels, when paying for more thinking actually returns more, and when ultracode earns its cost.

Tool calling

The real mechanics of how a model calls a tool, where tools break, and how to design ones it cannot get wrong.

Hooks

Rules that fire on their own before and after tool use. Deterministic guardrails at zero context cost.

Projects & structure

A repository Claude navigates instead of guesses at. Settings, permission modes, multiple environments.

Open-source frameworks

What exists, what's worth adopting, and how to judge one before you build your company on top of it.

Agents in production

Agents reading, deciding and executing against real company data. Monitoring, failure modes, retry, rollback.

Deployment

Laptop to production: environments, secrets, domains, shipping without breaking what already works.

Security & limits

What an agent must never be handed. Permission boundaries, secret handling, and drawing the line before something expensive happens rather than after.

What you leave with

A system running inside your own company, a skills library that is yours, and a written plan for extending both. Not a certificate, not a portfolio project.

The skills move with you

Claude CodeCodexCursorAntigravityHermesOpenClaw

Almost none of this is locked to one tool. Context engineering, MCP, tool design, agent architecture and cost control are how these systems work — not how one product works.

We build on Claude Code because it is the deepest implementation available today — not because it is the only one. What you learn here transfers to the others as they are.

Four days, and the week that holds them together

Sixteen hours, four hours a day. Two weekends, with seven days between them where the real learning happens — the same way tile is set: pieces first, then the bed that locks them.

Weekend one · Saturday

Foundation and context

Correct setup: permission modes, settings.json, wiring your first MCP servers
What belongs in context, what never does, and why output degrades as it fills
The CLAUDE.md hierarchy, and writing rules that actually hold
Structuring a repository Claude can move through fast
Token economics: caching, real cost per session in dinars, and how to cut it
Reasoning efforts — when raising the level returns more than it costs

You leave with a tuned environment and a real context file for your company.

Weekend one · Sunday

MCP and tools

Tool calling from the inside — the mechanics, and where tools actually break
Connecting existing MCP servers to your database, API and internal tools
Building your own MCP server from scratch
Evaluating open-source frameworks without locking yourself in
Secrets and permissions — what the agent sees, and what it must never see
Shipping to production: environments, variables, domain

You leave with a system wired to your real infrastructure, running.

The week between

Seven days where you work alone on your own system and hit problems nobody warned you about. This is not a gap in the programme — it is the part that does the work. Weekend two opens on what broke for you, not on prepared material.

A private WhatsApp group with me and the rest of the group
One specific assignment, due before weekend two

Weekend two · Saturday

Skills, subagents, workflows

Writing skills, testing them, and making them trigger when they should
Progressive disclosure — why a good skill cuts tokens and raises accuracy at once
Subagents: isolation, delegation, and the limits of what they can be trusted with
Teams and parallel agents across worktrees, without collisions
Workflows: fan-out, pipelines, deterministic verification stages
Ultracode — what it is, when to reach for it, what it costs
Hooks: guardrails that execute on their own, costing no context

You leave with your own skills library and a workflow running a real task.

Weekend two · Sunday

Systems in production

Agents reading, analysing and executing against your real company data
The automation layer: connecting systems and running on a schedule
When the agent gets it wrong: how you find out, and how you roll it back
Cost control over months, with real numbers rather than estimates
A one-to-one architecture review of what you have built
What to keep building after, and where the answers live

You leave with a system running in your company and a written plan to extend it.

What to bring, and what this will actually be like

Nothing below is softened. If any of it changes your mind, better now than on the Saturday.

Prepare before you arrive

A laptop you are allowed to install software on
An Anthropic account on an active plan
Node.js and git installed — I send you the steps
A GitHub account
One real problem in your company, written down in three lines
Access to whatever system that problem touches — read-only is enough

What to expect

Four hours of building each day. You will not be sitting through slides.
You will get stuck and irritated. That is the training working, not failing.
Claude Code costs money every month. You get the exact figures and how to lower them.
Skip the week between and weekend two is worth half of what it should be.
You do not walk out a software engineer. You walk out able to build and run systems for your own company.
Nothing here earns money on its own. These tools raise your output; they do not replace the work.

A small group, and I choose who is in it

Not scarcity for its own sake — past a small group I cannot sit with each person on their own system, which is the entire point of doing this in a room.

Right for you if

You run a company or a project with operations worth automating
You have used Claude or ChatGPT and know you are touching a tenth of it
You want to understand the machinery, not just operate the surface
You can commit four full days plus the week between them

Not for you if

You want a general introduction to artificial intelligence
You want a finished no-code solution handed to you
You are looking for a certificate to show an employer
You cannot attend all four days

Previous rooms

Raslem presenting to a room of business owners, a slide about token budgets projected behind him — A previous session with Algerian business owners

Raslem answering a question from the room during a workshop — Working through one participant's problem