SheepIT SHEEPIT SARL-S · Luxembourg Software architecture and development

Software architecture and development for systems that can’t afford to break. Including the AI ones.

SheepIT is a software company in Luxembourg, founded in 2022. I’m Emmanuel Mouton, and I run it. I have been designing and writing software since 2008, in banking, insurance, telecom and public institutions, and I am still the one writing the code. Today I also design and operate the AI agents that handle SheepIT’s own back office.

SheepIT SARL-S /Luxembourg /Est. 2022

01

Work

Four areas. The AI work is the most recent one. The engineering discipline underneath it is not.

a

AI agents and automation

Agents that do office work unattended: reading mail, filing documents, reconciling figures, reporting on a schedule. The first decision is which parts should not be an agent at all, because fixed logic belongs in a script and only judgement belongs to an agent. The rest is boundaries: what it may touch, what it hands back for approval, what it may spend.

b

Architecture for systems under constraint

Systems that stay correct under load, under audit, and under requirements that keep moving: domain modelling, service boundaries, integration, and the route off a legacy platform nobody can switch off. Security sits in the architecture, not in a layer added later: authorisation flows, single sign-on, secrets, encryption. Regulated contexts, where being wrong is expensive and slow to find.

c

Delivery with AI in the loop

I use AI inside my own engineering work, not only as something I build for clients: generating and reviewing code, arguing against my first design, writing the tests I would have talked myself out of. It makes an experienced engineer faster. It does not make an inexperienced one experienced, and it never decides anything hard to reverse.

d

Technical leadership, hands on

I have been a CTO, a team lead, and the person still writing the code, sometimes all three in the same month. Setting up engineering practice (delivery pipeline, review, testing, security), making architecture decisions hold once the pressure arrives, and coaching the people who have to live with them.

02

About me

Emmanuel Mouton

Emmanuel MoutonGérant, SheepIT SARL-S

SheepIT is a Luxembourg software company, founded in 2022, and I run it. I have been writing software since 2008, well before the company existed, mostly the kind that other people’s money, documents or networks depend on: banking and fintech, insurance, telecom, and European institutions. I work from Luxembourg, in French and in English.

The longest stretch was Expeto: from 2019 to 2025, architect and senior developer on a platform for private enterprise mobile networks. The domain model, the reporting pipeline, and single sign-on against each customer’s own identity provider. Since mid-2025 I have been at the Council of the European Union, on a workflow application most of its document teams rely on.

Before that, tokenised securities at Tokeny, and regulated back offices: ING Luxembourg, Société Générale, Pictet, Foyer. For fourteen months I was also the CTO of Docunify, acquired by Victor Buck Services in May 2019.

Two things I have not done, since the difference matters when you are deciding whether to write to me. I have never trained a model: my AI work is orchestration, tooling and control around models other people train. And I have never run an engineering organisation. The largest group I have led was ten people, and I was coaching them, not running a department.

What I keep coming back to is the same question in different clothes: which decisions in this system are hard to reverse, and are they written down anywhere. The AI work is interesting to me for exactly that reason. An agent with no boundary is a very fast way to make an irreversible decision.

03

Stack

What I actually use. Listed so you can tell in ten seconds whether we would be speaking the same language.

AI systems and agents

LLM orchestration, tool-calling agents, Model Context Protocol servers, retrieval on PostgreSQL with pgvector, evaluation gates, deterministic guardrails, budget and rate control

Back end

Java, Kotlin, Groovy, Scala, Spring and Spring Boot, Java EE, NodeJS, REST and messaging, domain-driven design, CQRS and event sourcing where the audit trail is the point, microservices where they earn their keep

Data

PostgreSQL, Oracle, MongoDB, schema and migration design, time-series processing, reporting and reconciliation flows, data retention under audit

Security and identity

OAuth2 and OpenID Connect, SAML, Keycloak, single sign-on against a customer's identity provider, secrets management, encryption at rest, threat modelling, remediation after penetration testing

Front end, when the work needs it

React, Angular, TypeScript. Enough to build and review a customer-facing interface end to end, not enough to call myself a front-end specialist

Delivery and runtime

Docker, Kubernetes, cloud and edge deployment, continuous integration and delivery pipelines, observability, release practice for systems that cannot take a maintenance window

Practice

Architecture decision records, code review, automated testing, threat modelling and secure development, technical coaching, working across distributed teams and time zones

04

Contact

Write to contact@sheepit.eu, or use the form. Either way it reaches the same inbox, and I answer it myself.