Akli Reguig · freelance software engineer · Paris

I build backends.
Lately, agents too.

Over ten years of Java, Spring and DevOps. These days, a lot of AI: MCP servers, skills and plugins for AI harnesses, shared memory for coding agents. Products that fix problems people actually have, built with the right tool for each job. Made for daily use, not for the demo.

See what I buildGet in touch
zsh · claude code
~/areguig.github.io ❯ claude
🪨 petit-poucet · 31 notes loaded (preferences + areguig.github.io)
> my homepage is outdated and not very sexy. fix it.
● memory_search("what does Akli build")
● Bash gh repo list areguig --limit 100
● Write src/app/page.tsx
● Bash npm run build
✓ Done. You are looking at it.
~/areguig.github.io ❯

01 / what I do

Solid backends, and agents that do real work.

why

Solve a real problem

If nobody is annoyed by the problem today, nobody will use the fix tomorrow. Start from the people, not from the stack.

how

Right tool, right job

Sometimes an agent. Sometimes a SQL query and a cron job. Being fashionable is not a reason.

always

Automate the boring parts

Tests, pipelines, deployments, and now the repetitive work agents can take. People should spend their time thinking.

01

Agentic engineering

The plumbing that lets AI agents do real work, whatever the harness: Claude Code, OpenCode, OMP, Pi, Copilot. MCP servers, skills, hooks, shared memory. Small, fast tools that fit the way a team already works.

AI harnessesMCPagent memorySpring AI
02

Backends that hold up

Java and Spring for over ten years. Clean APIs, event-driven services, data that stays correct. Nothing is trusted before it is tested, including code an agent wrote.

Java / SpringAPIsevent-driventest-first
03

DevOps, for real

The part I love. Continuous integration and delivery, containers, Kubernetes, the cloud. Pipelines so boring that nobody talks about them anymore. That is the goal.

CI/CDKubernetescontainersGitOps
04

Bringing the team along

A tool nobody uses is a demo. Coding dojos, pair and mob sessions, and agents fitted into the way a team already works, from the PO writing stories to QA.

dojospair / mobAI workflows

02 / selected work

Things I built because I needed them.

open source · featured

petit-poucet

Leaves pebbles so your coding agents find their way back.

One memory for every AI harness I switch between. Claude Code, OpenCode, Codex, Cursor and Copilot all start each session with the same rules and verified facts. Plain Markdown in git, readable in Obsidian, served over MCP by a small Rust binary.

RustMCPMarkdown + gitmacOS · Linux · Windows
$ curl -fsSL …/install.sh | sh
$ petit-poucet setup
  ✓ claude code   ✓ copilot
  ✓ codex         ✓ cursor
  ✓ opencode      ✓ antigravity

🪨 20 notes loaded
## Preferences (all repos)
- commit-rules: never push
- never-read-secrets

↳ The agent that rebuilt this site already knew me. That’s petit-poucet at work.

Lair

Coding agents you talk to out loud. Claude, Codex, Copilot or any ACP agent answers by voice, comes back when its work is done, and asks for your OK from the notch. All speech stays on the Mac.

SwiftACPMCPon-device voice

Refinery

Planning poker for backlog refinement, made for the shared screen. Votes stay secret until the reveal. When they diverge, the highest and lowest voters explain.

PocketBaserealtimeno accounts

scrawl

A scratchpad for AI agents. The agent publishes a single HTML page and gets a link to share. Reports, diagrams and prototypes stop dying in the terminal.

Cloudflare WorkersD1agents

Karate Studio

Karate as a service: write API tests in the browser, run them, read the report. Third rewrite. That tells you how much I like Karate.

Java 21QuarkusKarate

03 / day job

Renault, since 2019. Six chapters.

Freelance since 2017, with Renault’s digital teams since 2019. Very different products, same habits: tests first, automated pipelines, and building blocks other teams can reuse.

2019 →

Connected services

The backend behind a car-owner mobile app. Reactive, event-driven, in the cloud.

reactiveeventscloud
2019 → 2024

Vehicle engineering data

APIs that describe how vehicles are built, for the people who design and make them.

APIsdataperformance
2022 → 2026

EV charging

A charging platform, from the charger on the wall to the back office.

OCPProamingKubernetes
all along

Shared building blocks

Starters, scaffolds and libraries. Written once, used by many teams.

Spring BootsecurityCI/CD
all along

Testing as a habit

Automated tests at every level, running in every pipeline. Karate Studio was born here.

KarateAPI testingend-to-end
2025 →

AI, for everyone

Agents in production, assistants for teams that don’t code, and the MCP servers that connect them to everything else.

agentsMCPSpring AI
recent · 2026

An agent that investigates production

Ask why something broke, get an answer backed by evidence. Plus a memory that keeps only what it can prove.

agentsobservabilitymemory
recent · 2026

Feedback that sorts itself

Users talk, AI listens, and the noise turns into clear tickets.

embeddingsLLMclean architecture

04 / the lab

Experiments. Some ship, some teach me something.

I have a full-time mission. This is what happens on evenings and weekends, when an idea won't leave me alone. There are more than you'd think.

kagentTests that write themselves, and mend themselves when they break.
PodWhispererAsking a Kubernetes cluster what hurts, and getting a straight answer.
offline agentsCoding agents on open-weight models, with no cloud in the loop.
OCPP simulatorA car park full of make-believe chargers, talking to a real backend.
ChargePathLong trips in an electric car, without the range anxiety.
arcadeThe games of my childhood, rebuilt to find where vibe-coding breaks.

Most of these live in private repos. Ask me about them.

05 / now · october 2026

What I'm on right now.

  • Shipping Lair: coding agents you talk to, out loud.
  • Shipping petit-poucet: one memory shared by every AI harness.
  • Growing scrawl, so agents hand over a page instead of a wall of text.
  • Teaching an agent to investigate production, and to remember only what it can prove.
  • Bringing AI to teams that don’t write code.

06 / toolbox

What I reach for, most days.

Build

  • Java & Spring
  • TypeScript
  • Rust (agent-written, human-reviewed)
  • SQL & PostgreSQL
  • Event-driven design

Ship

  • Continuous integration & delivery
  • Containers & Kubernetes
  • Cloud (GCP, Cloudflare)
  • GitOps

AI

  • AI harnesses
  • MCP servers & agent memory
  • Spring AI
  • Open-weight models

Practice

  • Test-first
  • Pair & mob programming
  • Coding dojos
  • Small, frequent releases

off the keyboard

Self-hosting

A small server at home runs the house and whatever needs trying next. Still the best ops school there is.

off the keyboard

Retro gaming

A shelf of handhelds, old and new, and strong opinions about screens and emulators.

off the keyboard

Old cars

Fixed so they stay on the road. A car that keeps running is greener than a new one.

07 / contact

Got a backend to fix, or a team that wants agents to do more than autocomplete?

Send a message on LinkedIn. Every one gets read, and answered.

Message me on LinkedIngithub.com/areguig