Taking on new builds Based in Oran, Algeria · Remote Systems · Automation · AI

I design systems
that do the work.

I build AI workflows, automations, and small tools that turn slow, messy marketing operations into customer acquisition machines. Multi-agent pipelines, no-code automations, and bespoke AI skills, engineered to run with minimal hand-holding.

View the work → Get in touch ↗
20minTopic → publishable draft
25+Bespoke AI skills shipped
9Agents in one pipeline
3Platforms orchestrated

Selected Work

01 / Index
01

Multi-Agent Content Pipeline

An assembly line of specialized AI agents that takes a topic to a publishable draft in ~20 minutes. Each stage is gated by an independent verifier, with a final human checkpoint.

Multi-agentVerificationPrompt systems
02

Bespoke AI Skills Suite

25+ purpose-built AI skills and agents for reviewing, editing, planning, and creating long-form content. A reusable toolkit that standardizes quality across a whole team.

Agent designWorkflowQuality systems
03

Research & Data Automation

A no-code automation that assembles research and analytics data on a trigger (metrics, search insights, screenshots), then hands a clean package to the pipeline.

Make.comNo-codeData plumbing
04

In-Sheet Research Agents

AI agents that run live web research directly inside a spreadsheet, turning thousands of rows of raw entities into enriched, structured data without leaving the grid.

AI agentsEnrichmentScale
05

Teaching Claude to Design On-Brand

A design system encoded for an AI: I taught Claude to generate on-brand visual assets from our guidelines. Captured below as a case study, iteration by iteration.

Design opsFigmaFeatured ↓

Case Studies

02 / Deep Dives
Case 01 · Design Ops · Human-in-the-loop

Teaching Claude to design on-brand assets, from guidelines, not guesswork.

Most teams treat AI design as a slot machine. I treated it as a system. I distilled our brand guidelines into explicit, machine-readable rules: color logic, line weight, composition, layout patterns by asset type, then iterated with the model until it could match the reference. Below: the target on the left, what Claude produced on the right, and the honest trail of attempts underneath.

Reference target Reference
The reference I gave it
The target style and composition Claude had to match.
Final on-brand result Result ✓
What Claude produced
On-brand and reproducible on demand.

Iteration Log / the attempts in between

off-brand to on-brand
Attempt 01 01
First pass. Off-brand
Right idea, wrong language. Ignored line weight and color.
Attempt 02 02
Color corrected
Palette locked to the system. Composition still cluttered.
Attempt 03 03
Layout rules added
Asset-type patterns in. Lines finally thin and clean.
Attempt 04 04
Near miss
Almost there. Refined spacing and hierarchy in the prompt.
Case 02 · No-code Automation

Make.com scenarios that run the busywork.

I build no-code automations that wire our tools together and handle the repetitive work end to end: turning ideas into validated keywords, sizing up the competition, and keeping the Notion workspace tidy. Three of them below, each a live Make blueprint you can open.

Content idea to validated keyword list scenario 01

Content idea → validated keyword list

Takes a content idea from our Notion workspace and turns it into a keyword list with search volume data, so we can validate and prioritize what to write.

View blueprint ↗
SERP and Google AI Overview analysis scenario 02

SERP & Google AI Overview analysis

Analyzes the first page of Google results and the AI Overview for keywords we want to rank for, then writes the analysis straight into Notion to support our SEO research.

View blueprint ↗
One-click Notion property sync scenario 03

One-click Notion property sync

Copies every database property from one Notion page up to its parent page in a single click, removing a tedious manual housekeeping step.

View blueprint ↗

How I Work

03 / Capabilities
C-01

AI Workflow Architecture

Designing multi-agent systems with clear stage boundaries, checklists, and independent verifiers, so output is reliable, not lucky.

C-02

Prompt & Skill Engineering

Turning fuzzy human judgment into explicit, testable instructions. Reusable skills that hold quality steady across people and runs.

C-03

No-Code Automation

Make.com scenarios that pull, transform, and route data on triggers, gluing tools together into hands-off pipelines.

C-04

Tooling & APIs

Small, sharp utilities and APIs that solve one job well: batch processors and data resolvers built for scale.

C-05

Knowledge Architecture

Structuring information systems in Notion that teams and agents can both navigate. The backbone for everything above.

C-06

Design Ops with AI

Encoding brand and design rules into systems an AI can execute, for consistent visual output without a designer in every loop.

Operating Principles

04 / Approach
i.

Systems over heroics

A process that runs without me beats a brilliant one-off. I build for the second hundred outputs, not the first.

ii.

Verify, don't hope

Every stage gets an independent check. Trust comes from gates and rubrics, not from the model having a good day.

iii.

Human at the gate

Automation does the volume; a human makes the final call. The interesting work is designing where that line sits.

iv.

Ship, then sharpen

Get a working prototype in front of reality fast, then improve it against real failures instead of imagined ones.

Open to interesting problems

Let's build the
machine for it.

habib@benabdeslam.com
GitHub ↗ LinkedIn ↗