Python Stewardship Working Group

Attributes

Property	Value
Date Created	2024-11-01
Target End Date	2025-02-01
Epic	https://gitlab.com/groups/gitlab-org/-/epics/15580
Slack	#wg_python-stewardship (only accessible from within the company)
Google Doc	Agenda (https://docs.google.com/document/d/1gs-OrjjyfxQ3BDaKxOXcrMuUl3z1jmsxGmbdmEgIBF8/edit?tab=t.dma9z3zh8fwb) (only accessible from within the company)
Meeting Calendar	Calendar (https://calendar.google.com/calendar/u/0?cid=Y18xZWE4ZTViZWZiYmUzMDk0MDgzNGJiZWViMWY1NTFlODVjNWQ0NzQwZDc0MzJhMWQyMDkzOWQ4MzU0YjhkNjU3QGdyb3VwLmNhbGVuZGFyLmdvb2dsZS5jb20) (only accessible from within the company)

Goal

• Consolidate Python development within GitLab, providing confidence to developers on developing high-quality Python codebases.
• Simplify developer decision making by providing opininated guidelines on how to create, maintain and deploy Python services and libraries in GitLab.
• Define resources (courses, mentorship) for training developers that want or need to contribute to a Python codebase.

Context

While Python has been present at small scale for many years in the company, the evolution of our AI Infrastructure lead to Python becoming a critical part of our product. Yet, we haven’t consolidated a development culture around it.

As a result, developers are confused on how to write Python code, how to review Python code, and how to set up new functionality. Ruby developers also feel a lack of support when it comes to understanding and contributing to the codebase. One key difference between our Ruby codebase and our Python codebase is that the Python codebase is more distributed: instead of having a single repository we have multiple each responsible for an area.

Examples of existing Python codebases:

• AI Gateway https://gitlab.com/gitlab-org/modelops/applied-ml/code-suggestions/ai-assist/
• Prompt library https://gitlab.com/gitlab-org/modelops/ai-model-validation-and-research/ai-evaluation/prompt-library
• Duo Workflow service https://gitlab.com/gitlab-org/duo-workflow/duo-workflow-service

Key milestones

• 17.7 - Kick off of Working group
• 17.8 - Contributions to documentation
• 17.9 - Define templates for Python services, libraries, linting and utilities

Exit Criteria

Criteria	Progress	DRI	Start date	End date
A non-Python developer has guidance on how to understand and contribute to Python codebase
Guidelines and templates for Python codebase creation (as described in outcomes) have been merged
Code review and maintainership guidelines for Python codebase have been merged
Guidelines for deployment of each of the three application types

Outcomes

Documentation and process:

• Training for non-Python developers
• Development guidelines
  • Different types of repositories:
    • Python services (AIGW, workflow)
    • Python utilities (CEF)
    • libraries
  • Creation of codebase
  • Authentication
  • Testing guidelines
• Code review and maintainership guidelines
• Deployment guidelines
  • Python services (AIGW, workflow)
  • Python utilities (CEF)
  • libraries

Development experience:

• Extraction of common setup
  • Linters
  • Frameworks
  • Code Structure
  • CI for deployment

Roles and Responsibilities

Working Group Role	Person	Title
Executive Sponsor	Tim Zallmann	VP of Engineering
Facilitator	Eduardo Bonet	Staff Fullstack Engineer, Custom models
Member	Fred de Gier	Staff Fullstack Engineer, MLOps
Member	Alejandro Rodríguez	Senior Backend Engineer, AI Framework
Member	Alexander Chueshev	Staff ML Engineer, AI Framework
Member	Tan Le	Staff ML Engineer, Model Validation
Member	Stephan Rayner	Senior ML Engineer, Model Validation
Member	Tetiana Chupryna	Senior Backend Engineer, Duo Chat
Member	Mohamed Hamda	Senior Backend Engineer, Custom models
Member	Dylan Bernardi	Backend Engineer, Editor Extensions
Member	Shola Quadri	Associate Backend Engineer, Code Creation
Member	Vitali Tatarintev	Senior Backend Engineer, Code Creation

Non-goals

• Evolution of AI Gateway, although related to this WG proposal, is out of scope.
• Data Science use cases (such as code to develop a new ML model) differs substantially to system development, and merits their own set of guidelines.