Skip to main content

Reference Templates for AIPDS

Reference templates provide reusable patterns for common AI Product deployments.
They help ensure consistency, clarity, and compliance across different AI Product types.

Why Templates Matter​

  • Consistency → Standardized fields reduce ambiguity.
  • Speed → Teams can bootstrap AI Products quickly.
  • Governance → Ensures critical declarations (e.g., risk, compliance, monitoring) are not skipped.
  • Extensibility → Templates can be adapted for new AI Product categories.

Template: Large Language Model (LLM)​

aipds:
  product_name: "Enterprise LLM"
  version: "1.0.0"
  deployment_targets:
    - cloud:
        provider: "AWS"
        region: "us-east-1"
  packaging:
    - api:
        type: "REST"
        openapi_spec: "url-to-spec.yaml"
  inference:
    modes: ["online", "streaming"]
    latency: "<200ms per short request"
    throughput: "100 RPS per replica"
  monitoring:
    performance_metrics: ["accuracy", "latency"]
    drift_detection: true
    fairness_metrics: ["equalized_odds"]
  retraining:
    strategy: "triggered"
    thresholds: {"accuracy": "<90%"}
  security:
    auth: "OAuth2"
    rate_limits: "100 requests/min"
  governance:
    risk_category: "High"
    prohibited_uses: ["surveillance", "autonomous weapons"]
  economic_model:
    pricing: "per 1K tokens"
    sustainability: "0.4 kWh / 1K tokens"
  lifecycle_state: "Production"

Template: Vision Classification Model​

aipds:
  product_name: "Medical Imaging Classifier"
  version: "0.9.0"
  deployment_targets:
    - on-prem:
        hardware: "NVIDIA A100 GPU"
  packaging:
    - container: "dockerhub.io/org/med-vision:0.9.0"
  inference:
    modes: ["batch", "online"]
    latency: "50ms per image"
  monitoring:
    performance_metrics: ["precision", "recall"]
    explainability: "grad-cam"
  retraining:
    strategy: "periodic"
    frequency: "monthly"
  security:
    auth: "Mutual TLS"
    data_protection: "HIPAA compliant"
  governance:
    risk_category: "High"
    compliance: ["HIPAA", "FDA guidelines"]
  economic_model:
    pricing: "per scan processed"
  lifecycle_state: "Development"

Template: AI Agent​

aipds:
  product_name: "Customer Support Agent"
  version: "2.1.0"
  deployment_targets:
    - agent_runtime:
        framework: "LangChain"
  packaging:
    - sdk: "pip install support-agent"
  inference:
    modes: ["online", "hybrid"]
    concurrency: "500 sessions"
  monitoring:
    drift_detection: true
    abuse_detection: ["prompt injection"]
  retraining:
    strategy: "human-in-the-loop"
    governance_approval: "Customer Success Officer"
  security:
    auth: "OIDC"
    prohibited_tasks: ["financial advice"]
  governance:
    risk_category: "Limited"
    audit_logs: "retained 90 days"
  economic_model:
    pricing: "per conversation session"
  lifecycle_state: "Production"

Usage Guidance​

  • Templates should be customized per organization.
  • Governance, compliance, and sustainability must always be tailored.
  • New categories (e.g., multi-agent ecosystems, generative design tools) should have dedicated templates.

Summary​

  • Reference templates help teams start with strong defaults.
  • Each AI Product must fill in and adapt a template to its specific use case.
  • Templates evolve as the ecosystem expands — ensuring AIPDS remains future-proof.

Principle: A well-specified AI Product begins with a template, but always requires contextual adaptation.