Foundational Pipeline Build Reference: Stages 1–5
Operationalises Steps 2–4 of the deployment cycle as a sequential build checklist. Covers Git setup, Docker by Model Paradigm, CI construction, the Provenance Fork, and MLflow Model Registry with paradigm-specific artifact handling. The Cycle Owner works through each stage in order.
Part of: The Public-Sector AI Deployment Cycle: A From-Scratch Playbook for Zero-Maturity Teams
- Format
- DOC
- Access
- Pro
Reference document
Foundational Pipeline Build Reference: Stages 1–5
Operationalises Steps 2–4 of the deployment cycle as a sequential build checklist. Covers Git setup, Docker by Model Paradigm, CI construction, the Provenance Fork, and MLflow Model Registry with paradigm-specific artifact handling. The Cycle Owner works through each stage in order.
Document ID
DOC-0049
Category
General
Access tier
PRO
Date
31 August 2026
Stage 1 — Git Repository Structure Setup
Feeds into: Governance Gate 1 (Stage 6 — First Security Sign-Off) as foundational prerequisite.
Before any code is written, establish version control. This is the first concrete deliverable of the Foundational Pipeline.
Checklist
- 01Install Git on the deployment environment (on-prem server or cloud tenant).
- 02Create a central repository using a self-hosted Git server (Gitea, GitLab Community Edition, or equivalent).
- 03Initialise the repository with the standard structure below.
- 04Configure branch protection:
mainrequires at least one pull-request review before merge. - 05Add a
.gitignorethat excludesdata/raw/,mlruns/,.env, and any model artifact files exceeding 50 MB.
Standard repository structure:
ai-deployment/
├── .github/workflows/ # CI pipeline definitions (Stage 3)
├── docker/ # Dockerfiles by paradigm (Stage 2)
├── src/
│ ├── training/ # Training pipeline (in-house path only)
│ ├── inference/ # Serving or scoring code
│ └── preprocessing/ # Data preparation scripts
├── configs/ # Pipeline and model configurations
├── tests/ # Evaluation and validation tests
├── docs/ # Governance documentation
│ ├── model_cards/
│ └── governance_gates/
└── data/
└── .gitkeep # Raw data never committedImportant
Never commit raw datasets to Git, regardless of data sensitivity level. The Classification Intake from Stage 0 resolved your Four-Dimension Branching — regulated data requires encrypted external storage. Use DVC or a shared object store for data versioning.
Continue with a free account
This document is part of our Pro client library. A free account gives you access to all free resources only — to unlock the full library, become a client.
Free accounts unlock free resources only.
Get this document
Available to Pro clients. A free account covers free resources only.
Become a clientDocument info
- Format
- DOCPDF export
- Access tier
- Pro
- Category
- General
- Published
- 31 August 2026