Karl-Josef Hilger kj-hilger

Hi, I'm KJ 👋

I build resilient infrastructure for Agentic AI within governed, deterministic workflows.
My focus is on engineering scalable, audit-ready architectures for highly regulated industries where data sovereignty, compliance, and autonomous execution meet.

Top Skills: Agentic AI • Camunda 8 • Spring Boot • Kubernetes • Terraform
Engineering: Java • Spring Security • Python • CI/CD Pipelines (GitHub, Tekton, Maven, OWASP, Sonar)
Sovereign AI: LLM Inference (vLLM, LM Studio) • LLM Optimization (PagedAttention, KV-Cache Tuning, FP4–FP16 Quantization) • Multi-Arch Kubernetes (ARM64/x86 GPU-Passthrough)
Infrastructure: Linux • AWS (VPC, IAM) • Infrastructure as Code • GitOps (Helm, ArgoCD) • MLOps
Data: SQL • PostgreSQL • AWS RDS Aurora
Observability & Compliance: IT Observability (Grafana, AWS Cloudwatch) • Compliance as Code (DORA, KRITIS)
Frontend & UI: Client/Server communication • JavaScript foundation
AI Harness Engineering: • Agentic Tooling, MCP & RAG Integration • Self-Hosted Agent Sandboxes (OpenHands, Docker) • GitOps Integration
Methodology & Leadership: Scrumban • Trained ~150 in BPMN/DMN & ~20 in Camunda process development & ~15 in Agentic AI

Open for exchanges on Agentic AI & Sovereign Cloud architectures
Please visit my professional profiles for professional history and career details.

My professional role involves navigating the high-stakes complexities of making critical trade-offs between quality, velocity, and cost along with "political" discussions in a complex, large-scale enterprise landscape - these outputs remain confidential within corporate repositories.
In contrast, this public space serves as a distillation of that experience, showcasing a modular, full-stack engineering approach.
My repositories isolate specific technologies or explore organizational patterns at a meta-level to provide clear insights into my engineering DNA.
Built entirely from scratch, my repos serve as sandboxed environments for technology validation on my own sovereign hardware and as a means for clear communication with stakeholders.

While I rely on AWS Cloud, GitHub Copilot and Frontier LLMs in my professional day-to-day work, my private lab is built differently - optimized for rapid prototyping, technological validation, and data sovereignty.
Hardware: Desktop (16 GB GPU + 64 GB RAM) • NVIDIA Jetson Orin Nano (8 GB Unified Memory)
OS: Debian-based distributions
IDE & Plugins: IntelliJ IDEA • Maven Helper, GitToolBox • SpotBugs, TestMe, TestAutomation • Terraform & HCL • Kubernetes, Go Template • DevoxxGenie
Local AI Harness for Interactive Review: DevoxxGenie (IntelliJ) with MCP (fetch-url-mcp) • LM Studio (Gemma-4-12B-QAT)
Remote AI Harness for Autonomous Software Development: Self-Hosted Agent Sandboxes (OpenHands, Docker) • Self-Hosted vLLM (Qwen2.5-Coder-7B-Instruct-AWQ) with fallback to Cloud-Hosted OpenRouter (Qwen3-Coder-Next)

A lightweight, ephemeral DevOps lab for rapidly spinning up a complete Camunda 8 Agentic AI process application together with an LLM provider on NVIDIA Jetson Orin Nano.
Operate Camunda 8 with Tool Calls to a Local LLM within enterprise-style Kubernetes. Don't worry about internet connection at runtime, token costs, update friction, or API changes.
Use this project as a starting point to understand how all components of a self-managed enterprise setup interact and prototype further scenarios.
Optimized for the NVIDIA Jetson Orin Nano 8 GB, a compact, cost-effective, and widely available standard edge device that is easy to reset and reprovision. It enables quiet, low-energy operation and is perfectly suited for portable, on-site demonstrations in air-gapped environments.
Topics: agentic-ai ai-automation bpmn camunda camunda-8 devops edge-ai helm-chart java jetson kubernetes linux llm-orchestration local-llm ollama platform-engineering self-managed shell-script spring-boot

Impact: Awarded for Delivery Velocity at a leading European Financial Institution.
Streamlined production deployments by reducing manual documentation overhead and cross-platform data handling by up to 80%.
Multi-Source Integration: Orchestrates data from Jira (Agile PM), LeanIX (Metadata), Kubernetes (Deploy Strategies), Grafana (Metrics), and ServiceNow (ITSM) into a single source of truth.
Compliance-as-Code: Engineered a Python-based tool to automate DORA, ITSM, GitOps and internal PM requirements, ensuring 100% adherence to standards.
Topics: compliance-as-code compliance-automation confluence confluence-integration dora jira-integration kanban leanix-integration local-llm-integration servicenow-integration