Architecture that won't hold
Stack and architecture picked by AI, never validated: database, hosting, code structure, the foundations in short. You accept because it works. The day you realize it's the wrong fit, it's too late to turn back.
AI build coaching · Custom software AI can take off. I know how to land.
I fly alongside you while you build with AI, or I take the controls and land your software in production. Solid, secure, compliant, and 100% yours.
45,000+ flight hours
Home base: Messancy · Luxembourg, Belgium, France · EN / FR
01 / The reality
With AI, everyone builds. Little of it holds.
An app vibe-coded by a founder, a homegrown tool a team built on its own, subscriptions stacked for lack of anything better: it moves fast, then it jams. In my experience, six fragilities show up almost every time:
DIY security
API keys in plaintext, homemade auth, secrets in the codebase: in other words, your house keys left under the doormat. One known vulnerability and everything is exposed.
Performance and costs out of control
Heavy code, cascading queries, AI bills that climb without warning. The tool crawls, and it costs more than planned.
GDPR and AI Act never reviewed
Personal data flowing freely, implicit consent, sub-processors outside the EU, users who don't know they're talking to an AI. One audit and it's over.
Nothing gets checked
You fix one bug, you ship three new ones. Releases go out by hand, with no tests and no way to roll back. And nobody checks that the AI built into the tool answers correctly, or still does after a model change.
Nobody uses it
The tool impresses in the demo, but it doesn't hold up in day-to-day work: it isn't connected to your software, it doesn't follow how the team works, and nobody set out what it was supposed to deliver. Three months later, everyone is back to their spreadsheets.
02 / The flight plan
From the ground to production.
Whether you fly it yourself with me beside you, or I take the controls, every project follows the same flight plan. And it starts on the ground, not in the code.
Step 01
Understand
Before any code: how your teams work today, who does what, with which data and tools, where things get stuck, what must stay human. You leave with a written scope and success criteria.
Step 02
Take off
AI gets you going fast: prototype, first features, wired into your existing tools, shown to the people who will use them.
Step 03
Stay on course
Architecture, security, GDPR, AI Act, tests. What doesn't show in a demo, and decides whether it holds.
Step 04
Land
Production launch, automated deployments, monitoring. We measure real usage and adjust with user feedback. Your software runs, and it's yours.
03 / Working together
Two ways to work together.
Whether you want to stay hands-on or delegate, the goal is the same: solid, secure software, ready for real users. And either way, AI saves you time and money.
AI build coaching
You build. I make sure it holds.AI gets you moving fast, but it won't tell you when the architecture is about to give way, when an API key is lying around in the code, or when there's a GDPR gap. I join your team as an embedded AI engineer (a Forward Deployed Engineer) and part-time CTO: you stay in control, with a senior at your side for every important decision.
For you if
- you're building your product with AI and want to do it right
- you have an MVP that's starting to get real users
- you don't (yet) need a full-time CTO, or don't have the budget
- your teams are already building tools with AI, and they need to reach production properly
What you get
- Scoping to start (how you work, what you need to gain), then a technical assessment: code, architecture, security, GDPR, costs
- AI wired into your existing systems (ERP, CRM, email, files), within your architecture and your access rights
- Architecture and stack choices validated before you code
- Regular code and security reviews, plus tests that check what the AI answers
- The right AI tools for your team (Claude Code, Cursor…), chosen and configured, with Git, tests and automated deployment
- AI costs tracked and capped, for a predictable bill
Three intensities
Light
Standard
Intensive
How it works
I build, you guide meWe start with scoping (how your team works, what it needs to gain), then a technical assessment of what exists. Then regular video calls, on-site visits when useful, and an intensity that adapts to your pace. Along the way, I transfer the skills to your team: good practices, annotated reviews, AI tools used well. Minimum 3-month engagement.
End-to-end build
You describe. I deliver. You own it.I design and build your software from A to Z, from first mockup to production. I build with AI myself: timelines and cost drop sharply compared to traditional development, with a senior's rigour on top. On delivery, everything is yours: code, data, hosting. No forced subscription, no lock-in. That's what I call selfware: software built for you, and owned by you.
For you if
- you need custom business software and don't have time to build it
- you're juggling subscriptions and spreadsheets that don't fit how you work
- a team loses time on a manual process: hunting for information across tools, re-typing, chasing
- you have an AI prototype to turn into a real product
What you get
- Web and/or mobile software in production, secure and designed for GDPR compliance
- A clear fixed price, agreed together upfront
- Documented, transferable code another developer can pick up
- Tested AI features, wired into your existing software (ERP, CRM, email, files) and, if needed, your AI assistants
- Success criteria set at scoping, and checked after launch
- 14 days of fixes included, then optional monthly maintenance and upgrades
Typical projects
Business tool
Client portal
Mobile app
Automation
How it works
Build it for meWe start from your business: who does what, with which data, where it gets stuck. I give you a written scope and a fixed price, build in stages and regularly show you where things stand, ship to production, then adjust with your users' feedback. After that, I stay on for maintenance if you want.
Torn between the two? Let's take 30 minutes, free and with no commitment, to see what fits you. Book a call
04 / Work
Products that made it.
Before AI and with it: software taken all the way to production, solid, secure and ready to grow. From backend to mobile, from architecture to deployment.
Before Abecedia · co-founder
Docunify, from founding to acquisition
I co-founded Docunify, a Luxembourg B2B SaaS fintech.
- Co-founder, CTO and project lead
- FEDIL Start-up Innovation Prize 2018
- Acquired by Victor Buck Services, POST Luxembourg group
A product solid enough to win over a jury, business customers, and then a buyer.
The experience I bring to your projects
Develux · Fractional CTO
MGM Garage Management
A team had built their garage management software with AI: working in a demo, but not ready to onboard real paying customers.
- Full architecture: web app, admin back office and mobile app
- Modernized interface, secured payments and customer data
- Automated deployments, in minutes
A complete, reliable product in production, and a team that became autonomous.
AI build coaching · Embedded Fractional CTOVisit the site
Develux · Fractional CTO
Keep Focus Construction Platform
Software for the construction sector, functional but fragile and risky to evolve.
- Consolidated and secured technical foundations
- Fully rebuilt interface (web + 42-screen mobile app, in 6 languages)
- Safeguards to avoid nasty surprises in production
A solid, multilingual app, ready to grow with confidence.
AI build coaching + buildVisit the site
Side project · built with AI
Grimzoo
An app to write or dictate your memories, ask them questions, share them or turn them into a book.
- Designed, developed and shipped entirely on my own
- Artificial intelligence, security and data privacy
- From idea to live product
Concrete proof of what I can deliver alone, end to end.
End-to-end buildVisit the site
Also in production: at Develux, a Peppol e-invoicing gateway (sending and receiving through a certified access point), connected to My Garage Manager and Keep Focus. See the gateway
In progress: an HR tool, AI video analysis for football, a student-staff management app for a restaurant owner and, at Develux, InfoEcole (school management) and KeepFood (restaurants).
05 / The standard
The Abecedia standard.
A quality bar I hold end to end, and the right expertise brought in when your project needs it. That's what makes the difference between a prototype that runs and software you can entrust to real users.
Production-ready, continuously
Throughout the project, I hold the bar: code and architecture, security, tests, versioning, deployments, environments. No technical debt piling up behind your back.
Code reviewTests & CI/CDSecurity & GDPR
Verified AI, controlled costs
When your tool includes AI, I check that it answers correctly, and still does when the model changes. I also track its consumption, with caps, so the bill stays predictable.
AI testingCost trackingCaps
AI Act, by design
Your users know when they're talking to an AI, generated content is labelled, important decisions are logged and validated by a human. I handle the technical side of compliance from day one, instead of patching it in at the end.
TransparencyTraceabilityHuman oversight
Interfaces where humans stay in control
A good model isn't enough: you need a good interface. 20 years of senior frontend put to a new question, how humans validate, correct or take back control of what the AI does.
AI UXValidationSenior frontend
Independent validation, when needed
Based on project sensitivity, the code and technical choices can go through an independent validation, carried out by an external senior backend engineer. An expert eye on top of mine.
External senior backendSensitive projectsAt any stage
Web design & UX, when needed
When looks and user experience matter as much as the tech, I bring in an external designer (web designer, art director or UX designer): visual identity, web design, design system, UX audit.
External designerVisual identityUX audit
06 / About
Pierre Haegeli26 years in web development.
“I've spent my career coding, shipping, debugging in production. Today I help companies and project owners take AI from demo to production.”
Embedded AI engineer (Forward Deployed Engineer) · Fractional CTO. Senior full-stack developer, from interface to production, for startups, large companies and public institutions. Co-founder of Docunify (B2B fintech SaaS, FEDIL Prize 2018), acquired by the POST Luxembourg group.
Tools change, the traps stay the same. I code with AI every day, which lets me ship faster and for less. My job: pass that speed on to you, and spare you the traps AI doesn't avoid yet. A senior engineer who still codes, and takes the time to explain.
European institutionsArcelorMittalPOST Luxembourg Group
Download my CV (PDF)07 / Questions
The questions people ask me.
01Coaching or end-to-end build: how do I choose?
If you want to stay in control and build it yourself with AI, go for coaching: you code, I set the architecture, review, secure, and code the foundations when needed. If you have neither the time nor the wish to build, or if the tool is critical to your business, hand me the end-to-end build: I deliver, you own it. We can also start with one and switch to the other. That's exactly what the first call is for.
02How does coaching work?
We start with scoping (how your team works, what it needs to gain and how we'll measure it), then a technical assessment of what exists. Then we pick an intensity together: Light (mostly advice), Standard (half advice, half hands-on) or Intensive (mostly hands-on). We schedule regular video calls, and pair-code on hard points. You stay in control of your product, I code the foundations in parallel and transfer the skills to your team as we go. Minimum 3-month engagement, renewable, 30-day notice.
03How is pricing calculated?
No public rate card: every project has its own number. End-to-end build: a fixed price agreed after scoping, plus a monthly amount if you want maintenance. Coaching: a monthly amount based on the chosen intensity. We talk about it openly on the first call: you explain your context, I give you an honest estimate and a transparent framework before any commitment. No opaque package, no surprise.
04Why not just hire a senior dev directly?
A senior is expensive, takes 3-6 months to recruit, and is a risk if your product isn't stable yet. With me, you've got someone within days, on 3-month renewable terms, no payroll, and you can adjust intensity over time. The goal is precisely to make you self-sufficient, not to keep you dependent. The day you hire an in-house CTO, I hand off cleanly.
05Is my data secure?
NDA from the first call if needed. Read-only access to code and infrastructure, never to production data unless explicitly requested. And we discuss GDPR early: it's rarely the hardest part, but almost always neglected. Based on project sensitivity, an independent security review by an external senior backend engineer can be done before going to production (Abecedia standard).
06Does the AI Act apply to me?
If your tool has users talking to an AI or generates content, yes, at least in part: since August 2026, users must know they're talking to an AI, and generated content must be labelled. Some so-called "high-risk" uses, such as recruitment or credit scoring, will face much heavier obligations from December 2027. I handle the technical side from the design stage; for the legal analysis, your adviser remains essential.
07Which AI features can you build in?
The ones that serve your business: answering in plain language from your own documents and data (RAG), dictating instead of typing, answering out loud, reading a photo or a document, spotting what happens in a video, generating images, sorting, summarizing, pre-filling, drafting a checked first version. And automating a process: AI agents that fetch information from your tools (ERP, email, spreadsheets), via connectors such as MCP, and work through the steps for you, with human sign-off at the sensitive moments. Always with the same safeguards: verified answers, capped costs, transparency for users.
08Which AI should you use: a major provider, a European provider, or a model on your own server?
It depends on your data, your volumes and your constraints, and we decide it together. The major providers' models, used through an API, are the most capable way to start. A European provider or EU hosting meets data-location requirements. An open-source model on a dedicated server keeps everything in-house when confidentiality demands it. I compare quality, cost, confidentiality and compliance, then design the tool so the model can be swapped without starting over.
09Do you work internationally?
Yes. Based in Messancy (Belgium), I work in French and English, first in Luxembourg, Belgium and France, and remotely across European and North American time zones (UK, USA, Canada). Remote day to day (calls, written updates, code reviews), and on site with you when it helps.
10How does the first call go?
30 minutes over video. You tell me about your project, what's blocking you, what worries you. I tell you frankly what I see, what's urgent, what can wait. No commitment, and even if we don't end up working together you leave with clear directions.
11What is selfware?
Software built for one specific person or business, which they fully own: code, data, hosting. The opposite of a generic SaaS you rent, which evolves on the vendor's roadmap. AI has slashed the cost of building such tools; my job is to make sure yours holds up in production, is secure and compliant, and can keep evolving. The one limit: payroll, statutory accounting or medical records, where certified software remains the right choice.
08 / Contact
Ready to land?
Whether you want to build it yourself or hand me the project, let's talk. 30 minutes over video, no commitment, and you leave with a clear direction.
Or drop me a few lines.
I reply within 24 hours and we find a slot together.