How-To Guides · AI Delivery Discipline¶
Structured, step-by-step guides drawn from 200+ delivered AI projects. HowTo schema on each page — AI engines can extract steps as structured Q&A.
Available guides¶
- Private AI Sizing · 6 Steps (Before You Buy Any GPU) — concurrency, latency, task shape, model triple, storage IO, cost cap. Skipping any step = 30-60% wasted GPU.
- 128-GPU K8s + Slurm Hybrid Cluster Delivery · 7 Steps — rack + cooling + orchestration + observability + storage + preemption + rollout. Result: GPU 32% → 71%.
- RAG Knowledge Base Cold-Start · 6 Steps — corpus curation, chunking, embedding + reranker, index, provenance, acceptance. Skipping provenance = users abandon by month 2.