software-architecture.aiSoftware that still holds up in two years.
From custom SaaS to AI agents and knowledge assistants, all the way to the prototype you built yourself and that now needs to carry weight. With architecture, tests and maintainability from day one. Built solo, with the care the work demands.
Sound familiar?
When your tool shapes your business instead of the other way around, it's time for something that fits your brand and your processes.
Your service inbox has become the bottleneck.
Every question gets answered by hand. Standard helpdesks don't fix it, because they don't know your brand voice and don't have your founder knowledge.
Standard SaaS doesn't fit your brand.
Mail tools suggest generic drafts. Shop plugins force their workflows on you. You work around the software instead of with it, and the brand voice dilutes along the way.
Custom software was always a different league for your size.
Traditional agencies were never an honest option for solo founders and small brands. So you built around the software instead of having it built for you.
Why custom over standard.
Standard software covers eighty percent. The remaining twenty percent are often what makes your brand specific: mailings that sound like you, service that answers in your founder voice instead of from a generic FAQ. For exactly that twenty percent I build software that fits your brand and your stack.
Clean architecture decisions before the first line of code. The difference between an app that grinds after three months and one that grows with your brand. I think the structure ahead, you stay on the business.
What does that look like in practice? Read the article
What I actually build.
Custom software solutions, end to end: from custom SaaS through AI agents and knowledge assistants to the prototype I take over and make production-ready. All from one hand.
Custom SaaS solutions
Your own software-as-a-service, tailored to your process: login, multiple users, billing, dashboard. Built to grow with your customers instead of forcing you into someone else's workflows.
AI agents and agentic systems
Agents that take on multi-step tasks: research, operate tools, chain steps. AI automation for your business processes, with clear boundaries, so control stays with you and the machine handles the routine.
Chatbots and knowledge assistants
AI chatbots for your business or website that answer from your own knowledge instead of generic platitudes. Retrieval-augmented (RAG) over your documents, products, FAQs. In your voice, with traceable sources.
MCP servers and integrations
AI integration into your existing systems: MCP servers give AI secure access to your tools and data, so shop, CRM, mail and helpdesk finally flow together.
Web apps, portals and dashboards
Need a customer portal, an internal tool or a reporting dashboard built? Clean web applications with modern architecture, built for your team and your customers.
Brand voice engines
Systems that write in your brand voice instead of the language-model default. Tuned to your founder knowledge, embedded where your stack allows: mail tool, shop, or editor.
Among my clients
Two projects I'm allowed to show publicly. Others run under NDA and stay unnamed here.
Get your time back
AI automation for your business processes
Most hours don't go into the important work, they go into the same routine around it. That's exactly what can be automated. The goal is simple: the dull routine goes to the machine, and the hours stay with the work that's actually yours.
- 1
The routine
The same task that eats your day: typing quotes, sorting requests, moving data from A to B.
- 2
The automation
An AI system takes on the dull steps. With clear boundaries and you as the final instance that decides.
- 3
The time
You get the hours back. For customers, for ideas, for the why. The part no machine can do.
What typically gets automated
- Prepare quotes and invoices instead of retyping them every time
- Pre-sort requests and generate reply drafts in your voice
- Keep data in sync between shop, CRM and mail tool, without copy-paste
Building AI agents
From chatbot to an agent that takes over work
An agent that genuinely takes work off your plate is several parts working together. I assemble them to fit the systems your business already runs on, so the routine runs on its own and the decisions stay with you.
For the technically curious: how an agent is built
Under the hood it's a handful of building blocks, and this is one way to put them together: LangChain for the components and RAG, n8n or LangGraph for orchestration, the Anthropic Agent SDK as the runtime, and MCP servers as the connector to your systems. It's one possible combination, not the only one. There are other orchestration frameworks and other agent SDKs too; treat this as a representative cross-section. Which pieces actually fit depends on your case. Here's how this example slots together, click through it, or we walk it live in a meeting.
How you build an AI agent: what it's made of
Flip the orchestration layer (n8n or LangGraph) and click through the layers. You see what each piece does and where it fits in.
Same layer, two routes. Both can hold for your approval. You pick by need, not by fashion.
One stack you assemble.
The pieces don't compete across layers, they complement each other. Two things stay flexible: orchestrate low-code with n8n or as a code graph with LangGraph (flip it, the rest stays), and an Agent SDK is optional, often LangChain on its own is enough. Click any layer to see what it does.
How an agent works, step by step
The same stack, now in action. Walk through the five steps and see which piece kicks in when. Pick between two typical cases from SMB everyday life.
A quote request lands in your inbox. n8n spots it and starts the agent.
The agent takes the routine. You keep judgement and the relationship.
The pattern stays the same, whether it's a quote, an inquiry or an invoice: the agent takes the repetitive routine, the decision and the contact with the human stay with you. These are exactly the stacks I build, tailored to the tools your business already runs on.
You already started
Your prototype isn't throwaway code. It's the specification.
What you build with AI over a weekend tells me more about your idea in ten minutes than a ten-page spec.
Vibe coding gets you to the start line. The prototype runs, you have your first users. Then it starts to wobble: bugs pile up, the API breaks on real data, and nobody knows how secure the whole thing is. That's exactly where I come in.
- RunsThe prototype stands, the idea is proven, the first users are here.
- WobblesBugs multiply, the API breaks on real data, security is a question mark.
- CarriesTaken over, hardened, cleanly rebuilt. Load-bearing for the next customers.
Your prototype is the specification, in runnable form. That's exactly my starting point: get the bugs under control, harden the error handling, assess the architecture honestly and rebuild it so it carries growth.
What that means concretely is a case-by-case call. Sometimes we keep the components that already hold and build cleanly around them. Sometimes a clean rebuild on what your prototype proves is the more honest path. Which one it is I tell you after the first look, not before.
You don't start from zero. You start where your prototype stopped carrying.
No pitch. Repair or clean rebuild? If you don't need me yet, I'll tell you.
Stack and integrations
Stack-close means I build where your business already runs, instead of forcing a new system on you.
- Web and app: React, Next.js, TypeScript, Tailwind
- AI systems: production LLM applications, RAG with embeddings and vector stores, brand voice tuning
- Agentic patterns: orchestration with n8n or LangGraph, building blocks and RAG with LangChain, the Anthropic Agent SDK as runtime, MCP servers for tool integration, maintainable for SMB everyday use
- Integrations: shop platforms, mail tools, helpdesks, CRMs, REST and GraphQL APIs
- Data: PostgreSQL, vector databases, event streams
- Operations: Docker, automated deployment, EU-hosted where it makes sense
Typical projects
- Custom SaaS with login, billing and dashboard for a growing team
- AI agent that takes on multi-step routine and leaves the decision to you
- Knowledge assistant that answers from your documents, with traceable sources
- MCP server that securely brings shop, CRM and mail tool together for AI
- Customer portal, internal tool or reporting dashboard for small teams
How it runs
Describe the project
We discuss requirements, goals and constraints. As peers.
Concept and architecture
A technical concept with clear architecture, schedule and a transparent cost estimate.
Agile delivery
Development in short iterations with regular demos. You see where we are at any time. After the first milestone you decide. If it doesn't convince you, you get your money back.
Handover and support
Clean documentation, deployment, walkthrough. After that I'm available for support and further development.
Optional
Grow together
On request I stay on after go-live: proactive optimisation, a monthly briefing, the security that the solution doesn't go stale but grows with you.
Without a tech head of your own
Your technical head, without hiring a CTO
Sometimes you don't need a finished project, but someone who carries the technical decisions with you: architecture, build-or-buy, security, an honest read on agency quotes and tool promises. Fractional CTO, outsourced CTO, technical sparring partner: a technical head on call, without paying a full-time salary or turning into the developer yourself. The way in is the 60-minute walkthrough, where you work out with me what you actually need.
Architecture quality, priced for SMBs.
AI-generated code has a bad reputation. Understandable. What gets clicked together without architectural understanding leaves marks. So I don't let it click things together. I steer it with the same architectural judgement I'd bring without it.
I've built my own development processes that integrate AI strategically into every step: from requirements analysis through architecture to testing. Architecture decisions, code reviews and quality assurance stay fully with me. The AI accelerates the build.
The result: software at a quality level that used to require enterprise budgets, at prices realistic for small companies.
The whole story behind it
Custom software was just for the big players. Now it's for you.How AI software development makes custom software affordable
What's behind the hype, why quality standards still hold and how custom software adds up for small companies.
Read the article→
Who's behind it
Over a decade of software architecture: from enterprise systems to lean solutions for small companies. The methods of large projects, without the overhead.
As a freelancer I know the reality of the people I work with: limited budgets, tight windows, no in-house dev team. Every solution has to work under those conditions.
“Good architecture isn't a luxury. It's the reason software still extends after two years.”
Freelance architect vs. in-house team
Now you know who's behind it. The remaining question: does an external architect make more sense than an in-house solution?
With a freelance architect
- Flexibly available exactly when you need me
- No recruiting, no overhead, no onboarding
- Instantly available expertise in software architecture and AI
- Tailored precisely to your project
- Transparent billing without fixed costs
In-house
- Lengthy recruiting for specialised profiles
- Fixed costs regardless of project workload
- Ramp-up time for architecture and technologies
- Additional management and administrative overhead
- Less flexibility with changing requirements
Both approaches have their place. For project-based architecture expertise, the freelance approach is often the more efficient path.
Does this fit your project?
So we both know from the start whether it fits.
Right for you
- You want professional software without needing an enterprise budget
- Your project is too unique for standard SaaS, but a traditional agency would be too expensive
- You value short paths: one point of contact from concept to deployment
- You want software that's cleanly built and can evolve over time
- You're looking for someone who thinks along, not just executes requirements
Not the right fit
- Your budget is below €4,000, where clean engineering with tests and architecture doesn't fit
- Your project needs a team of ten or more developers in parallel
- You need 24/7 managed hosting with SLA guarantees
- You're looking for a pure template website without custom logic
Happy to recommend the right partners for these topics. If you're unsure whether your project fits the corridor, let's still talk briefly.
Frequent questions
Sounds like you?
Describe your situation. I'll tell you honestly what can come of it.