FREN
CVGet in touch
Anas Abbassi
Get in touch
Download CVGitHub (opens in a new tab)LinkedIn (opens in a new tab)WhatsApp (opens in a new tab)
FREN
Casablanca · UTC+1

COLLECTION ’26

Anas Abbassi, agentic delivery engineer in Casablanca

I design and ship production applications by orchestrating coding agents: specification, parallelisation, review and hardening.

Specify, parallelise, verify again. The code goes to production.

View workDownload CVGet in touch

CasablancaUTC+1Available for permanent & freelance

  • 2.5+years in production
  • 3independent products
  • 5case studies
  • FR · EN · ARworking languages

01 · Selected work

Selected work

Four recent products, from client SaaS to an enterprise suite. Each card opens a case study covering the problem, the decisions and what was built.

All work

02 · Stack

The tools behind the method, not a logo wall.

Every node carries a delivery decision: an interface talking to the right store, a module served by CI, local memory that stays on disk.

Hover a node to see the projects that use it.

STACK / MAP

2026

STACK / MAPThe technologies I use in production.
LaravelPHPNode.jsMCPTypeScriptReactNext.jsElectronSupabaseSQLiteDockerGitLab CICloudflare

Jobmiko · Next.js · Supabase · Cloudflare

Backend
LaravelPHPNode.jsMCP
Frontend
TypeScriptReactNext.jsElectron
Data & infra
SupabaseSQLiteDockerGitLab CICloudflare

The technologies I use in production.The three families of the stack, at a glance.

03 · Services

How I deliver

Four steps, one chain: the brief becomes a specification, agents execute, review decides, production holds.

  1. Scoping and specification

    A brief becomes an executable specification: contracts, acceptance criteria, edge cases, verification plan.

    Deliverablesspec and contracts · acceptance criteria · breakdown into verifiable tasksHow I workThis is what makes agents useful: a spec they can follow and that I can refuse.

  2. Agent orchestration

    Several agents in parallel on disjoint scopes, coordinated through MCP, with project memory and review at each step.

    Deliverablesmulti-agent pipelines · dedicated MCP servers · persistent project memoryHow I workSupraCortex and SupraSpace come from this work: the tools I use, I build.

  3. Review and hardening

    Generated code is not shipped as it comes. Architecture, edge cases, security and data integrity are reviewed and tested.

    Deliverablesarchitecture review · security and data integrity · test suite and CIHow I workThe leverage is not the agent, it is the verification.

  4. Shipping to production

    CI/CD, environments, observability and handover. The deliverable is a system that runs and that another team can pick up.

    DeliverablesGitLab CI/CD pipelines · Docker containers · handover documentationHow I workSpeed only counts if it holds in production.

04 · Lab

Explorations

Six short experiments, from agent memory to delivery tooling.

View on GitHub (opens in a new tab)
Deterministic memoryJSONL by default, SQLite/FTS5 as an option, lexical retrieval.View case studyWorkflow in natural languageSpecialised agents produce a validated n8n JSON workflow.GitHubDelivery chainAn existing suite, a CI and environments that hold up under load.View case study
Local-first storageIndexedDB first; Supabase or Docker as an option.View case studyReal PTYsnode-pty and xterm.js, a grid from 1 to 6 panes.View case studyThe dependency mapWhat talks to what, in production, end to end.All work

05 · Notes

Notes

All notes
  1. Specify before you parallelise: why an agent without acceptance criteria produces unverifiable code.An agent is only reliable when its output can be refused. Without written acceptance criteria, every review turns into an opinion.
  2. Project memory: why I wrote SupraCortex rather than pasting context into every session.Pasting context by hand does not scale. SupraCortex keeps a local memory that agents query over MCP and that points back to its sources.
  3. Review as the bottleneck: what I check systematically in generated code.Verification becomes the point of contention, so it is what has to be tooled. In SEMS, CI and the Docker environments set what I can ship.

06 · About

I ship with agents, and I answer for the code produced.

I design and ship production applications by orchestrating coding agents. The work runs in four stages: specify, orchestrate, verify, ship. Agents speed up execution; the specification and the review stay mine.

The method comes from the tooling I built for myself: SupraCortex, local-first project memory for coding agents; SupraSpace, a local command centre for coding CLIs; FlowFactory, an experiment in generating n8n workflows.

The base remains production: the Procure-to-Pay module of SEMS at E-Solution since August 2024, with Laravel, GitLab CI/CD and Docker, after an internship on fleet-management submodules from February to July 2024. I use Laravel, Node.js or React when they are the right tool, never as an argument in themselves.

Method

Specify, orchestrate, verify, ship.

  1. Specify

    Contracts, acceptance criteria and edge cases are written before the first line. The spec serves both the agents and the review.

  2. Orchestrate

    Agents work in parallel on disjoint scopes, with shared project memory and a review at every step.

  3. Verify

    Architecture, security and data integrity are reviewed, and tests pass in CI before every integration.

  4. Ship

    CI/CD pipelines, Docker environments and handover documentation, so another team can carry on.

Why this is different

Why this is different

  1. One spec that both agents and humans follow.
  2. Agents parallelised on disjoint scopes.
  3. Every merge is reviewed and tested.
Specification
Contracts · acceptance criteria · edge cases
Orchestration
Claude Code · Codex · OpenCode · MCP
Review
Architecture · security · tests · CI
Shipping
GitLab CI/CD · Docker · observability

StackLaravel · PHP · Node.js · React · Next.js · TypeScript · Tailwind · Supabase · SQLite · Docker · GitLab CI

Learn more

Based in
Casablanca
Role
Agentic delivery engineer
Company
E-Solution
Education
EHEI
Languages
FR · EN · AR

07 · Contact

A project, a role, an idea?

Tell me the context, the problem and the deadline. I reply in French or English, from Casablanca.

Send an email

abbassianas43@gmail.com

Elsewhere

GitHub (opens in a new tab)LinkedIn (opens in a new tab)

Availablepermanent & freelanceUTC+1No form, no tracking: one email is enough.