Skip to content

Pilot deployment options

This guide shows how to pilot deployment options in a reliable, repeatable way.

Who this is for

  • Developers and platform engineers planning a pilot environment.
  • Buyers who want to understand what “running it” entails.

What you will get

  • A concrete set of pilot deployment shapes
  • What each option proves (and what it doesn’t)
  • A recommended progression from pilot to production

Option A: Local / developer sandbox (fastest)

Use when you want to prove:

  • API integration and request/response semantics
  • Tenancy isolation and headers/claims handling
  • Exposure log creation for later evaluation

Start with:

Option B: Single-tenant pilot environment (staging)

Use when you want to prove:

  • End-to-end integration from your app to RecSys in a networked environment
  • Observability, basic SRE runbooks, and rollback procedures
  • A real(istic) traffic shadow or small production slice

Recommended building blocks:

Option C: Production-like (artifacts + manifest + scheduled pipelines)

Use when you want to prove:

  • Daily refresh of signals with guardrails
  • Atomic publish/rollback and freshness SLOs
  • A repeatable evaluation workflow

Start with:

  1. Local sandbox → proves integration semantics cheaply.
  2. Staging pilot → proves ops readiness with real integration constraints.
  3. Production-like → proves long-term operating model (pipelines, ship/rollback, evaluation).