AI-generated
We build what I use myself. Architecture decisions, code, integration into your system, tests, handover. With synthetic data, in weeks. Typically 80–200 fixed-price hours. All documented.
Production-ready code with tests. Multi-agent setup when needed. Full integration into your systems (APIs, databases, third-party tools). Live deployment incl. monitoring. Handover training for your team. Documentation that stays maintainable.
Where your project data runs is decided in the initial call – before the first commit. Provider choice is part of the tech-stack planning; DPA templates are provided for projects involving personal data.
| Provider | Hosting | GDPR compliance | Use case |
|---|---|---|---|
| OpenAI Enterprise | USA, stateless API mode (no input persistence) | Standard DPA under GDPR Art. 28 + additional Data Processing Addendum (Enterprise tier) | Internal research, texts without PII, code reviews |
| Microsoft Azure OpenAI (EU region) | EU data centers (Netherlands, Germany, Sweden) | EU data storage, GDPR Art. 44 compliant, no third-country transfer | Highly sensitive data, regulated industries, finance and healthcare |
| Anthropic Claude | USA, stateless API (standard) / EU options via AWS Bedrock | Standard DPA, inputs not used for training (default) | Complex reasoning tasks, code generation, multi-agent workflows |
| On-Prem / Self-Hosted (Llama, Mistral, open source) | Your infrastructure – private cloud or own GPU servers | Maximum control, no third-party data transfer | Highest sensitivity, regulatory requirements, attorney-client privilege |
OpenAI Enterprise
Microsoft Azure OpenAI (EU region)
Anthropic Claude
On-Prem / Self-Hosted (Llama, Mistral, open source)
Provider choice is part of the initial conversation. DPA templates are provided for projects with personal data. Not legal advice – detail review in your compliance context.
80–200 fixed-price hours
Kick-off → prototype (fast) → validation → handover. Fixed price, confirmed in the intro call.
Concrete toolset, no vendor theatre. Six building blocks that appear in every project.
Editor and agent in one. Own OSS plugin (session-orchestrator) for structured multi-wave sessions with quality gates.
Next.js, TypeScript, Tailwind, Supabase for web. Python with uv for agent pipelines. Swift when mobile is needed.
Vercel AI SDK, Anthropic SDK, ConnectRPC are introduced only when pipelines need to run in parallel. No vendor drops for the vendor's sake.
TypeScript strict, ESLint, Playwright for UI, Pytest for Python. What is not green does not enter the main branch.
GitLab primary, GitHub as mirror, Conventional Commits. Every architecture decision is traceable in the repo.
Vercel for web, Docker for self-hosting. Sentry for production errors, clear escalation on incidents.
Client project onboarding takes four to eight weeks at most companies. With three standard tools, the first productive answers come in minutes. Reproducible, not theoretical.
Claude Project: set name, system prompt ("Reply only using sources from the uploaded documents."), and team access.
Three PDF documents (tender, technical specification, framework agreement) via drag-and-drop. Summary appears automatically.
CSV with line items and market prices. First calculation as a bar chart with variance commentary.
Every new document, request, and meeting record lands in the container and is linked with context. After one week: a fully searchable project history with sources.
Every answer in client interactions has a source reference, no statements without evidence.
Project onboarding time: from weeks to days, because the container carries the learning curve, not the individual.
Handover during leave or role changes without knowledge loss, the container is the documentation.
Scalable to all knowledge-intensive processes: tenders, supplier evaluations, HR onboarding, compliance reviews.
The tools themselves are trivial. The difference comes from one person per area who establishes the workflow. Two to six weeks for a pilot project. Start with an area where the recurring task is measurable.
Which tech? Database? Hosting? Auth? LLM provider? We settle these decisions before the first commit lands.
Fast iterations: build MVP features, validate with test data, incorporate feedback. You see progress every 2–3 days.
Write tests, cover edge cases, remove performance bottlenecks. Real data, real load. All green.
Deployment into your production. Monitoring set up. All levers explained. Your team can continue developing the solution independently.
Clear answers before you sign.
Every sprint commit goes directly into your repository. No vendor lock-in, no withholding.
Open-source dependencies under MIT/Apache 2.0. Proprietary modules listed individually.
Questions after go-live? Included. Beyond that: optional maintenance contract on hourly corridor.
ADRs, prompt docs and test coverage so your team can keep building after handover.
Handover and continuity can be set as a contract term: code, repo, and documentation are handover-ready from day one – your team is never dependent on my availability, during or after the build. We set availability and response times in the contract.
Not for clients, but for my own workflows. One is public on the Mac App Store, two are internal and honestly still in progress. This is what “builder” looks like in practice.
The agent proposes the full cut, I approve, ffmpeg renders. Transcript in, clean cut out: filler words, pauses and false starts drop away. The footage stays on my Mac, only the transcript text goes to the cut-brain.
Product page
My Instagram tool for my own brands. Posts are template-rendered: the text is a fixed layer, zero hallucinated copy. AI only for optional backgrounds, behind a vision judge that rejects images containing text. Learning loop adapts to real engagement metrics.
My macOS dictation tool, public on the App Store. Hold a hotkey, speak, release: clean text lands wherever your cursor is. New in the direct channel: Meeting Mode transcribes meetings locally, produces minutes and action items. Everything on-device, no cloud.
walkaitalkie.comAutonomous agent with prompt-injection defence (Soft-SGR + 5-layer hardening). Scoring based on observable side effects. 3-h build window, eval against 104 tasks: 79 of 104 points · 2nd place: 65.
View case studyA dashboard that bundles bookkeeping, invoicing, FinanzOnline submissions and time tracking, so SME owners spend less time on administrative overhead and more on the business. Open beta at buchhaltgenie.at. Features: invoices, recurring invoices and delivery notes · create customers, suppliers and products, also directly from chat · bank import with AI-driven categorisation · capture receipts by photo, AI extracts date/amount/VAT · validation against Austrian tax law (UStG/BAO) · FinanzOnline export via XML · Sophie, chat assistant for Austrian accounting · smart insights and cashflow forecast.
View case studyCoordination of an international webshop relaunch for a distribution group: 11 countries, 16 websites, 8 webshops migrated to a unified platform base. Role: IT and e-commerce project manager, phased country-by-country rollout to minimise risk.
View case studyChannel
On my YouTube channel I regularly show how I build agentically and via vibecoding: filmed openly, not staged. Not a project-specific showcase, but an honest look at the way of working behind these builds. Below, my latest video plays automatically.
Fixed prices by scope – as a reference before the initial call:
For larger builds (80–200 hours): hourly corridor €140–180, all prices net, excl. VAT.
Concrete pricing discussed in the initial call after scope clarification.
How a corporate-scale initiative works with one solo builder, without false promises.
Instead of one hard-to-steer mega-contract, I break an initiative into separately signed-off pilots. Each pilot has a fixed scope, one outcome, and a sign-off point before the next one starts.
You have one point of contact who builds, documents, and hands over, no rotating team and no pass-the-parcel. Who delivers what by when is written down before every pilot.
If a workstream exceeds what one person can reasonably deliver, I refer vetted partners from my network, individually and transparently. There is no standing team that I pass off as my own.
Even at larger scope, one person stays accountable, reachable, and familiar with your code in detail. The code sits in your repo from day one, standards and decisions are documented, and for a short-term absence there is a defined hand-over path, identical to the safeguard on the About page. Pace and parallel workstreams are agreed after scoping in the initial call.
30-minute intro call. Afterwards you know whether it fits.
