Purpose and Scope

This reference serves Step 8 (Stage 10 — Post-Deployment Monitoring with Prometheus/Grafana) in the deployment cycle. It provides concrete, copy-deployable configurations for the Cycle Owner who has just completed Stage 9 (Extending CI to CD for Production Release) and now needs a live monitoring dashboard before the system can be considered operational.

The monitoring scope branches by Model Paradigm, consistent with the playbook's terminology:

• Traditional ML — monitor model drift and data quality
• GenAI — monitor API latency, log prompts and responses, and implement content safety alerts

Prerequisites

Before applying this reference, the Cycle Owner should have:

1. 01A running Prometheus instance (installed as part of infrastructure provisioning in Stage 7)
2. 02Grafana installed and connected to Prometheus as a data source
3. 03A deployed model endpoint (traditional ML scoring service or GenAI inference API) exposing metrics on a /metrics path
4. 04Completed the Foundational Pipeline (Stages 1–3) so container orchestration is in place

Shared Metrics Configuration

Regardless of Model Paradigm, every deployment exposes these baseline metrics. Add the following scrape target to prometheus.yml:

scrape_configs:
  - job_name: 'ai-model-endpoint'
    scrape_interval: 15s
    static_configs:
      - targets: ['localhost:8000']
    metrics_path: /metrics