Tech
Your problem first. Then the stack.
To me, technology is a means to an end: I dig in and deliver — performance-driven, efficient, documented. Whatever problem shows up: the solution is already one step closer here.
From problem to solution
»We need a web platform that lasts.«
Clean PHP architecture: Laravel or Symfony when their ecosystem helps — my own engine when control and leanness matter. Database, auth and i18n thought through from day one.
- PHP
- Laravel / Symfony
- devgine
- MySQL / PostgreSQL
»Our internal tool costs more time than it saves.«
Lean internal tools with a short path to value: fast UI, automation, integration with what's already there — instead of a six-month project.
- PHP
- Python
- Bootstrap
- APIs
»Is our system up right now? No idea.«
Monitoring from the ground up: DNS, TCP and TTFB latency, uptime history, status page — no third-party providers. Your data stays with you.
- Custom monitoring
- Status page
- Alerts
»Our servers are looked after by … nobody, really.«
Set up, harden and permanently maintain Ubuntu servers — deployments, updates, backups. Just like the Behnke support server that has been running under my responsibility for years.
- Ubuntu VPS
- Docker
- Bash
»Nobody understands our documentation.«
Documentation platforms that actually get read: multilingual, accessible, searchable, with PDF export. Proven with a wiki in three languages.
- i18n DE/EN/FR
- Accessibility
- PDF export
»We're supposed to "do something with AI".«
Use case first, then the model: Claude, GPT, Gemini or local via Ollama. I build workflows that deliver measurably — and tell you honestly where AI adds nothing.
- Claude Code
- GPT
- Gemini
- Ollama
»Website's live — but Google can't find us.«
An SEO foundation instead of tricks: structured data, a clean meta architecture, load times you don't notice. Get found without an ad budget.
- JSON-LD
- Meta / OG
- Performance
»Community or game project with its own tech?«
From Minecraft plugins and mods to my own 2D engine: tech that carries the fun instead of slowing it down — operations included.
- Java
- Spigot API
- Forge
- libGDX
The toolbox
Development
- PHP
- Laravel · Symfony · own engine (devgine) · PSR standards
- Java
- Spring · libGDX · Spigot API · Forge · Maven / Gradle
- Frontend
- JavaScript · React · Bootstrap · vanilla where it's enough
- Graphics
- OpenGL · libGDX — 2D rendering, own engine
- More
- Python · C/C++ · Node.js
Data & operations
- Databases
- MySQL · PostgreSQL · SQLite — models, optimization, migrations
- Servers
- Set up, harden and permanently maintain Ubuntu VPS
- Deployment
- Docker · Bash · Git-based workflows
- Monitoring
- DNS, TCP, TTFB latency · uptime — self-built
Quality & process
- Analysis first
- Figure out exactly what's needed — then build
- Agile
- Scrum · short iterations · honest estimates
- Code quality
- PHPStan max · tests as a merge condition · PSR-12
- Documentation
- Fully documented source code — others can step in at any time
Web foundation
- Security
- CSRF · secure sessions · parameterized SQL · disclosure processes
- i18n & a11y
- DE/EN/FR · OpenDyslexic · screen ruler · contrast
- SEO
- JSON-LD · meta architecture · OG / Twitter cards
- Performance
- No framework bloat · no third-party requests
AI — the right model for the right job
- Claude & Claude Code
- Pairing while building: refactorings, reviews, entire features in dialog — strong when the task is clearly scoped.
- GPT
- Research, writing, quick prototypes, brainstorming — opens up thinking space before decisions are made.
- Google Gemini
- Multimodal tasks and large contexts — when images, documents and code need to be looked at together.
- Local models (Ollama)
- When data must not leave the house — deliberately smaller, but fully under control.
- The principle
- No model is a team member: every line gets reviewed, the value gets measured — and AI gets switched off where it costs quality.
- Scientific foundation
- My bachelor's thesis systematically compared eight LLMs for business process modeling — custom evaluation framework, custom prompting strategies, real benchmarks. I don't believe in the AI hype: I measured it.
The stack in icons