DAG Cards - AI Developer Tools Tool

Overview

DAG Cards is a small, template-driven utility that generates static HTML “DAG cards” for Metaflow pipelines by combining structural DAG metadata with recent run data (for example from Weights & Biases) and rendering the result via Jinja templates. The project is presented by Jacopo Tagliabue and collaborators as an extension of the model-card idea to full ML pipelines, focusing on internal, team-facing documentation that surfaces owners, step-level details, recent metrics and behavioral-test outputs in a single, shareable HTML page. According to the repository README, the tool reads the Metaflow client representation of a Flow, pulls run and artifact information (the example integrates W&B), prettifies JSONs, and fills a front-end template (Chart.js is used for charts) to produce a stand-alone card file. ([github.com](https://github.com/jacopotagliabue/dag-card-is-the-new-model-card)) The implementation is intentionally an MVP: it demonstrates concrete examples (a sample training_flow and a card_builder.py driver), is built for Python 3.6 compatibility, and is structured so the Jinja template and service adapters can be extended to other experiment-tracking or artifact stores. The authors also published a companion paper describing the concept of DAG Cards and their motivation in a data-centric AI workflow. The repo is MIT-licensed and meant primarily as a reference / starting point for teams that want automatic, code-driven pipeline documentation. ([github.com](https://github.com/jacopotagliabue/dag-card-is-the-new-model-card))

GitHub Statistics

• Stars: 30
• Forks: 7
• Contributors: 1
• License: MIT
• Primary Language: Python
• Last Updated: 2021-12-07T13:29:22Z

Repository activity and community signals are modest: the GitHub project shows 30 stars and 7 forks, a small number of contributors, and the last recorded commit in the main branch is December 7, 2021. The project has no published releases and no open issues or pull requests at the time of inspection. These facts indicate a stable but low-activity research/demo repository rather than an actively-maintained production library; teams should be prepared to fork and adapt the code for their needs. ([github.com](https://github.com/jacopotagliabue/dag-card-is-the-new-model-card))

Installation

Install via pip:

git clone https://github.com/jacopotagliabue/dag-card-is-the-new-model-card.git

cd dag-card-is-the-new-model-card/src || cd dag-card-is-the-new-model-card

python3 -m venv .venv && source .venv/bin/activate

pip install -r requirements.txt

Create a local .env file from the provided template and populate required keys (Metaflow profile, W&B API key)

Run a sample DAG (example): METAFLOW_PROFILE=my_profile python training_flow.py run

Build a card from the sample DAG: METAFLOW_PROFILE=my_profile python card_builder.py

Key Features

• Generates standalone HTML ‘DAG cards’ using Jinja templates and embedded Chart.js visualizations.
• Extracts DAG structure from Metaflow to render step-level collapsible details and visual DAG maps.
• Fetches recent run metadata and metrics (example integration: Weights & Biases) for model sections.
• Includes Owners, Overview, Model metrics (latest K runs) and behavioral test sections in cards.
• Outputs a static HTML file suitable for emailing, Slack, or archiving alongside run artifacts.
• Modular template structure—easy to extend adapters or add other tracking services or config options.

Community

The project is an open-source, MIT-licensed demo/research artifact with limited ongoing maintenance: 30 stars and 7 forks indicate community interest, but the last commit was December 7, 2021 and there are no active issues or releases. The repo and accompanying blog/paper provide reference implementations and design rationale; contributors should expect to fork and extend the code for production use. For details and to contribute, see the GitHub repo and the authors' paper and blog post. ([github.com](https://github.com/jacopotagliabue/dag-card-is-the-new-model-card))

Key Information

• Category: Developer Tools
• Type: AI Developer Tools Tool