Meet Konrad: The AI SRE That Runs My Infrastructure - and Can Never Act Alone
A self-hosted AI SRE that restarts deployments, drains nodes and patches hosts. A pattern-matched quarantine, not the model, decides what may run.
Marcus Pichler
CTO THINKING. BUILDER MENTALITY.
From Edge to Cloud to AI.
More than 25 years in enterprise IT: retail platforms, connected mobility, IoT, cloud and Kubernetes. My focus now is AI.
I design, build and operate AI platforms: retrieval, agents, observability. Self-hosted models where the data has to stay in-house, hosted APIs where they fit better.
The assistant on this site runs on my own hardware: Gemma 4 12B via Ollama on a single RTX 3060, with retrieval over Qdrant.
Key Strengths
AI · Edge · Cloud · DevOps, built over 25+ years
AI Architecture & LLMOps
AI platforms end to end: inference, retrieval, agent orchestration, tracing and cost control, on self-hosted models and hosted APIs.
Edge & Cloud Architecture
AWS Certified Solutions Architect with hands-on experience from sensors and gateways through to Kubernetes.
Platform Engineering
Infrastructure as Code, GitOps and CI/CD automation, with observability and security designed in from the start.
Recent Experience
Enterprise IT since 2001
Nov 2025 - Present
IT Architect
Land Salzburg, Austria
Architecting and implementing an on-premise AI platform on NVIDIA B300, alongside end-to-end IoT solutions from the device layer to cloud connectivity.
Oct 2022 - Oct 2025
IoT Solution Architect
PIERER Innovation, Salzburg
Orchestrated IoT project implementations for connected motorcycle platforms, including technical specifications and integration processes.
2001 - 2021
Various Senior Roles
SPAR ICS, Salzburg
20 years progressing from IT support and POS technology to Enterprise Architect, Innovation Lab Technical Lead and Senior Product Lead.
Featured Projects
Systems I have designed, built and operate
ALOIS - Agentic AI Platform
pichler.devSelf-hosted multi-agent platform on LangGraph and LangMem. Seventeen personas on one runtime, each with its own role, tools and long-term memory.
RAG Chatbot on Self-Hosted Infrastructure
pichler.devThe assistant on this site: Gemma 4 12B on a single RTX 3060, hybrid retrieval over Qdrant, running entirely on self-hosted infrastructure.
AI Red Team Assessment
pichler.devOWASP LLM Top 10:2025 security assessment against the production RAG chatbot on this site: white-box audit combined with live attacks from inside and outside the network.
Latest from the Blog
Notes on AI infrastructure, agents and operations
Monitoring and SIEM for a Homelab on Single-Node K3s
Prometheus, Grafana, Loki, Wazuh, CrowdSec, Falco and a honeypot on single-node K3s: 163 endpoints, 22 agents, 84 dashboards.
Building a Self-Hosted RAG Chatbot From Scratch
How I built a self-hosted RAG chatbot for my portfolio website - from GPU inference to OWASP security, running entirely on my own hardware.