From automation scripts
to enterprise AI systems
First automation. First client.
Started with Python automation scripts for small businesses. Quickly realized the bottleneck wasn't code — it was architecture. Built the first RAG prototype connecting 3,000+ documents to a custom retrieval layer. Client cut document search time from 40 minutes to 8 seconds.
Computer Vision for industrial use
AirConductor — gesture-controlled presentation software using MediaPipe + Python. Replaced the Bluetooth clicker entirely. This opened the door to DACH B2B industrial clients needing vision systems for quality control and production monitoring.
Patented software system
NextGen Pathways — a zero-hallucination AI trainer for call-center operators — was registered as patented software. It answers only from verified training material, or honestly refuses. Proof that the architecture holds up where accuracy matters more than fluency.
DennisCraft AI Studio — autonomous infrastructure
Founded DennisCraft AI Studio. Built OpenClaw — a 24/7 AI orchestrator managing 9 projects in parallel via Telegram. Three AI agents coordinate through a shared file-system pipeline. Zero VC. Zero outside capital. All revenue from B2B client projects.
What I actually build
Autonomous AI Agents
Multi-step reasoning pipelines that act without human intervention. LangGraph, custom tool-calling, retry logic, state management. Production-grade orchestration.
RAG Pipelines
Retrieval-Augmented Generation with semantic search over private documents. Qdrant/Pinecone vector stores, chunking strategies, reranking, hybrid search.
Computer Vision
Real-time object detection, gesture recognition, industrial quality control. YOLOv8, MediaPipe, OpenCV. Runs on edge hardware, not just cloud.
Agentic Workflows
LLM-driven business process automation. Email classification, document extraction, CRM integration, automated reporting. End-to-end pipelines.
AI Infrastructure
Production deployment on Hetzner/AWS/GCP. Docker, monitoring, logging, cost optimization. Runs 24/7 without babysitting.
LLM Fine-Tuning
Custom model adaptation for domain-specific use cases. LoRA/QLoRA fine-tuning, PEFT, evaluation frameworks. Deploy locally with Ollama.
How I design AI systems
Tools I trust in production
Verified milestones
Let's work together
I work with DACH companies that need AI systems that actually function in production — not demos, not prototypes. If you have a real problem, I have a real solution.