Technical library
Whitepapers.
Working-operator research on running frontier AI in production: model selection, inference economics, and the honest line between build, rent, and call an API.
Published
The papers
Two papers published, more in progress. Each is a worked, sourced analysis, not a vendor pitch.
Self-Hosting Open-Weight Frontier Models for Production Workflows
Which model, what hardware, what throughput, and what ROI, with GLM-5.2 and DeepSeek V4 as worked examples. 19 sections, 14 charts, and 51 cited sources, from footprint sizing to OpenRouter resale and EU AI Act exposure.
Laszlo Adam Toth · Budapest, EURead the paper →Exit Strategies When AI Subscriptions Tighten
What to do when Anthropic or OpenAI change the caps: a provider-independent continuity plan for engineering-critical AI. 16 sections, 3 charts, and 37 cited sources, from the anchor-workload cost math to gateways, managed-private routes, and the honest line on self-hosting.
Laszlo Adam Toth · Budapest, EURead the paper →
