PRD — AI/ML Product Manager

Overview

Draft and validate PRDs for AI/ML products: model behaviour, evaluation, guardrails, ethics, data requirements.

Example Conversation

You: I need a PRD for our new recommendation model. We need evaluation criteria, guardrails for harmful content, and data requirements.

Agent: I’ll use the AI/ML PRD skill: load best-practices and sections (Use case & model behaviour, Evaluation, Guardrails & safety, Data requirements), draft from template, then run validate_prd_structure.

Agent: Created prd-recommendation-model.md. Validation PASS.

What the Tools Validate

validate_prd_structure checks the PRD for required sections (Overview, Goals, Use case & model behaviour, Evaluation & metrics, Guardrails & safety, Data requirements, Success metrics, NFRs) and unfilled placeholders. Output: PASS / NEEDS_REVISION / FAIL.

Output Excerpt

AI/ML PRD (excerpt):

## Use case & model behaviour

- **What the model does:** Recommends next item from user history and context; ranking model
- **Failure modes:** Cold start; low diversity; inappropriate content

## Evaluation & metrics

- **Quality metrics:** NDCG@10, diversity, coverage
- **Acceptance criteria:** NDCG@10 &gt; 0.35; no regressions on diversity

## Guardrails & safety

- **Boundaries:** Exclude adult, violence; content policy filter
- **Human-in-the-loop:** Flagged items reviewed within 24h

Getting Started

Add your context (use case, metrics, guardrails) to the project workspace, then activate the PRD — AI/ML Product Manager skill. The agent will guide you through drafting and validating with validate_prd_structure.