Advisory engagements for teams that need a second opinion or a concrete plan. The deliverable is a document — an assessment, an architecture, a roadmap — not a training session. Flexible scope, billed by engagement. Prefer hands-on training on your own infrastructure? That's workshops.
One expert, a written deliverable, no training. You leave with a document you can hand to your team or your procurement department.
1–2 weeks · remote or on-site
A review of your existing infrastructure — Kubernetes, storage, networking, observability — with a written report identifying gaps, risks, and a prioritized target architecture. You get a document you can act on or hand to your team.
1–2 weeks · remote
Model selection, GPU sizing, RAG architecture, and cost modeling for running LLMs on your own hardware. Includes a build-vs-buy analysis and a deployment plan for Ollama, LiteLLM, and Dify. Based on the self-hosted AI guide and the Dify guide.
1–2 weeks · remote or on-site
Setup and tuning of privacy-centric AI infrastructure on the NVIDIA DGX Spark and DGX Station lineup — from a single DGX Spark (64 GB, shipping now) to an eight-unit Spark cluster with 1 TB of unified memory, or a DGX Station serving 30+ users. vLLM inference, RAG on your own documents, model selection, and a cost model — your data never leaves your building. Agent operations follow the NVIDIA OpenShell reference architecture: sandboxed agents with supervisor-checked policy enforcement, not prompt-away-able guardrails. Based on the self-hosted AI guide.
2–4 weeks · remote or on-site
A migration roadmap from SaaS tenants to a self-hosted stack — k3s, S3 backup, identity, email — with risk assessment and a phased rollout plan. Based on the k3s guide and the S3 backup guide.
1–2 weeks · remote or on-site
A hardening review against C5:2026, NIST CSF 2.0, and ISO 27001 controls — gap analysis, remediation priorities, and audit-ready documentation. Based on the compliance frameworks article.
Hands-on workshops are separate: fixed scope, fixed price, your team building working systems on your infrastructure — Kubernetes, self-hosted AI, observability, n8n, and more.
| Service | Price | Description |
|---|---|---|
| Discovery call | €0 | 30 minutes to discuss your setup and assess fit. No commitment. |
| Consulting engagement | By arrangement day rate or project | Advisory work above — assessment, architecture, roadmap. Billed by engagement or day. |
Tobias Weiß builds and operates research IT infrastructure at the University of Marburg — GPU clusters and AI/XR platforms used in daily teaching and research — and publishes production guides for self-hosted stacks on this site. Everything offered here is technology he runs himself first.
Write via the contact page or email hello@tobias-weiss.org. Every engagement can start with a free 30-minute call to check fit — describe your setup, and you get an honest assessment of what makes sense and what doesn't.