Use this structure when documenting an AI model, API or production deployment. It helps other builders compare capability, cost and operational risk.
Workload
- Model and provider
- Input and output types
- Expected traffic, context size and latency target
- Quality and safety requirements
- Data residency or compliance constraints
Architecture
Describe authentication, routing, retries, rate limits, caching, observability and fallback behavior. For self-hosted models, include hardware, quantization, inference engine and concurrency settings.
Evaluation
Use a representative test set and document accuracy, latency, throughput and cost. Check prompt injection, sensitive-data handling and failure behavior instead of relying only on a public benchmark.
Operations
Protect API keys, set budgets and alerts, record model changes and maintain a rollback path. State commercial-use and data-training terms relevant to the deployment.
