Model Card Demo¶

This demo walks through the Model Card generation demo -- an interactive demonstration that generates Model Cards and Datasheets for FCC personas, workflows, and categories.

Table of Contents¶

Introduction and Prerequisites
Launching the Demo
Single Persona Card
Risk Category Assignment
Batch Generation
Datasheet Generation
Markdown and JSON Output

Introduction and Prerequisites¶

System Requirements¶

Python 3.10+ with FCC installed (pip install -e ".[dev]")
No API key required

What This Demo Shows¶

The Model Card Demo demonstrates how to generate structured documentation for every FCC entity -- personas, workflows, and categories -- following the Mitchell et al. 2019 and Gebru et al. 2021 standards.

Launching the Demo¶

fcc demo run model-card

Or run directly:

from fcc.demos.model_card_demo import run_demo
run_demo()

Single Persona Card¶

The demo generates a model card for the Research Curator (RC) persona and displays key fields:

Model Card: Research Crafter
  Type: persona
  Version: 0.9.0
  Owner: INFORMATION COLLECTIVE, LLC
  Intended use: FCC Find phase persona for Research Crafter tasks.
  Risk category: None
  R.I.S.C.E.A.R.: Role: ..., Archetype: ..., Style: ...

Risk Category Assignment¶

When a constitution is available, the demo shows risk classification:

Persona: DGS (Data Governance Steward)
  Hard-stop rules: 4
  Risk category: HIGH

Persona: RAE (Responsible AI Ethicist)
  Hard-stop rules: 3
  Risk category: HIGH

Persona: RC (Research Crafter)
  Hard-stop rules: 0
  Risk category: None

The mapping follows EU AI Act risk tiers: - 3+ hard-stops = HIGH - Any mandatory patterns = LIMITED - Only preferred = MINIMAL

Batch Generation¶

The demo generates cards for the full registry:

Generating model cards for all entities...
  Persona cards: 102
  Workflow cards: 7
  Category cards: 20
  Total: 129

Datasheet Generation¶

The demo produces a datasheet for the persona dataset:

Datasheet: FCC Persona Dataset
  Composition: 102 persona YAML definitions, each with 10-component
    R.I.S.C.E.A.R. spec, discernment matrix, design target factors,
    and persona dimensions.
  Purpose: Configure AI agent personas for FCC workflow execution.
  Creator: INFORMATION COLLECTIVE, LLC

Markdown and JSON Output¶

The demo renders a sample card in both Markdown and JSON format:

---
title: "Research Crafter"
type: persona
version: 0.9.0
---

# Research Crafter

## Overview
- **Type:** persona
- **Version:** 0.9.0
...

Tips¶

Use generator.batch_render(cards, "output/", fmt="markdown") to write all cards to disk
Include model card generation in CI for automatic documentation updates
Review risk categories whenever constitutions are modified