Projects

AI at work and at home

I don't just use AI professionally. The most interesting solutions often emerge in the evening at the kitchen table — for family, the club, the neighborhood. Here's a selection.

Professional Projects

Automate customer service

Phone Assistant for Appointments

The problem: Customers wait on hold, staff spend time on routine requests. The solution: An AI that takes calls, understands the request, and books appointments directly in the system. Built in an enterprise environment, transferable to any business with appointment scheduling.

Details

Understands natural language — no menu navigation
Books appointments directly in existing systems
High success rate for standard requests
Available 24/7, no staffing costs

Feedback loop for autonomous agents

DevProcess Toolkit

Autonomous AI agents need structured feedback to improve. Manual testing doesn't scale. This system captures feedback automatically, categorizes and prioritizes — and fixes clear bugs completely autonomously. Open source, can be self-hosted, SDKs for all platforms.

Details

Agent-in-the-loop: Feedback directly into the dev process
Autonomous bug fixes for clear errors
Self-hosted possible, full data control
Open source with SDKs for Web, iOS, Android, React Native

GitHub

GenAI backend for an automotive virtual assistant

MB.GPT

Spring 2023 — one of the first GenAI projects in the German automotive industry, live before ChatGPT went mainstream. A GenAI backend for an automotive virtual assistant: extensive knowledge base for technical queries. Intelligent routing, topic detection, LLM-based response generation. Natural language database queries — comparisons across model lines, technical specs, all possible.

Details

Extensive knowledge base + vehicle database
Language processing: topic detection, Q&A, relevance ranking
Natural language database queries for technical questions
Responses enriched with images, videos, warnings, and links

Enterprise Virtual Assistant

Virtual Assistant (Automotive)

A personalized assistant for the entire customer journey — from product consultation through appointment booking to after-sales. As product manager, responsible for strategy and evolution. Combination of chatbot and generative AI with modular architecture: answering questions, searching manuals, booking appointments.

Details

All channels: WebChat, WhatsApp, App, In-Car, Call Center Support
Modular architecture with AI-based classification and orchestration
Multiple AI providers depending on use case
Deployed globally, next generation in pilot

Technical books as audiobooks — accessible to everyone

Agent Chronicles

Antonio Gulli's book on AI agents is brilliant — but dense. Not everyone has the time or background to work through 400 pages of technical literature. I wanted to multiply this knowledge: make it accessible without watering it down. The solution: A multi-agent system that transforms technical books into understandable audiobook episodes.

Details

12 specialized AI agents as a production team
From technical book analysis to finished audiobook in hours
ACX-compliant audio quality with ElevenLabs
Multiplies expert knowledge without the author as bottleneck

Real-time dashboard for voice assistants

VoiceBot Analytics

Voice assistants produce data. But which calls fail — and why? This dashboard shows live metrics, automatically classifies calls (human, voicemail, noise), and alerts on anomalies. 8 specialized analysis agents determine root causes. Real-time updates, seamless integration.

Details

Live metrics: calls, success rate, connection errors
Automatic call classification (human, voicemail, noise)
8 analysis agents for automatic root cause detection
Seamless integration for call metrics

Multi-Model Dev Platform — Claude Code rebuild without vendor lock-in

Forge

29 tools, 131 tests, model router for Anthropic, Google, OpenAI and local models. Autonomous development platform with cross-device support, quality engine and bootstrapping architecture.

Details

Multi-Model Router
29 Claude Code Tools
Cross-Device Sessions
Cost Tracking

Skill chains, quality gates and a learning loop that never stops

Autonomous Dev Team

56 skills, 6 hook types, KB agent as semantic knowledge store. Risk-based autonomy levels, circuit breakers, workflow state machine. From 10 manual triggers per feature to 2–3.

Details

56 Chained Skills
KB-Agent Knowledge Store
8-Level Autonomy
Circuit Breaker

Personal & Social Projects

Learning companion for school presentations

GFS Tutor

My daughter had a school presentation coming up — and I didn't have time to help every evening. The solution: An AI tutor that guides her through the entire process. Topic research, outline, presentation practice. Age-appropriate explanations, more patient than any human could be.

Details

Age-appropriate explanations, adapted to grade level
Structured research with source verification
Presentation practice with feedback
Time management until deadline

43 kids, multiple coaches, one AI assistant

My Training Planner

Youth football training with 43 kids. Doing it right — following DFB training philosophy, with age-appropriate game formats and the right mix of technique and free play — easily takes an hour of preparation. Between a full-time job and family, that hour doesn't exist. So I built an assistant that knows my team's context and creates didactically sound sessions.

Details

DFB-compliant: fun, intensity, repetition
Complete training session in 30 seconds
Knows team, focus areas and season planning
Foundation for KICo — the app for all coaches (Coming soon)

When a Swabian does the math himself

My Solarteur and His Team

Nearly 100,000 euros for a smart home? Not with me. I did all the electrical work in my house myself. When the solar quotes came in — same story. So I built a team of specialized AI agents: system integrator, structural engineer, tax advisor, solar consultant. Each with a clear role, all with my project knowledge base.

Details

Role-based multi-agent team for PV/ESS planning
26 markdown files as knowledge base
Energy storage system with Home Assistant and Tibber
Everything verified two and three times — agents love to hallucinate

When a gnome achieves more than any conversation

Nisse

My son is autistic. Processing the day is hard for him. Explanations bounce off, role-playing doesn't help. Then came Christmas — and with it Nisse, a gnome who writes letters. Behind it: an AI prompt that picks up the day's situation and wraps it in a story that actually lands.

Details

Personalized letters based on the day's events
Age-appropriate and adapted to development level
Interactive: my son writes back
Not a therapist replacement, but a quiet helper

My first book — coming 2026

The World Has 1000 Colors

A book about perspectives. About how the world isn't black and white, but shimmers in a thousand colors — if you look closely. About perception, diversity, and why we so often talk past each other. Coming 2026.

Details

Reflections on perception and perspective
Stories from everyday life and work
Illustrated with own photographs
Release date: 2026

Connected home, self-built

Smart Home

Not an off-the-shelf system, but self-planned and implemented: Proxmox server in the basement, KNX wiring throughout the house, Home Assistant as the hub, Alexa in every ceiling. All local, all under control, all expandable.

Details

Proxmox cluster with VMs and containers
KNX bus for lights, blinds, heating
Home Assistant as central control
Voice control in every room