Back to Projects

AI PM Toolkit

Planned

Generate PRDs, roadmaps, and user stories from a product brief using AI.

FastAPI OpenAI API HTMX Markdown Export PDF Export
GitHub Live Demo
❌

The Problem

PMs spend significant time writing formulaic documents (PRDs, roadmaps, user stories) but get stuck on structure, completeness, and clarity. Ad-hoc ChatGPT prompts produce mediocre results because they lack structure and rigor.

💡

The Approach

Build an AI toolkit with carefully engineered prompts that generate high-quality, structured PM artifacts. Provide three specialized modes: PRD Generator, Roadmap Builder, and Story Writer. Make output editable and exportable.

✅

The Solution

A web-based AI PM Toolkit that takes a product brief and generates professional PRDs, phased roadmaps, and user stories with acceptance criteria. All output is structured, editable, and exportable to Markdown or PDF.

What

An AI-powered toolkit that generates structured PM artifacts (PRDs, roadmaps, user stories) from a product brief.

Three Modes:

  1. PRD Generator — Input: product name, target user, problem, market context, constraints, goals

  2. Output: Problem statement, target user + JTBD, success metrics, MVP scope, technical considerations, risks, open questions

  3. Roadmap Builder — Input: product vision and timeline constraints

  4. Output: Phased roadmap (Now/Next/Later or quarterly), milestones, dependencies, success criteria per phase

  5. Story Writer — Input: features from the roadmap

  6. Output: User stories with acceptance criteria, edge cases, technical notes, grouped by epic

Additional Features: - Inline markdown editor for outputs - Export to Markdown (.md) or PDF - Save and browse generation history - Copy-to-clipboard for each artifact

Why

The Problem: Most PMs use ChatGPT with ad-hoc prompts and get inconsistent results. PRDs are either too vague or bloated. Roadmaps lack clear dependencies. User stories miss edge cases.

Why This Toolkit Matters: - Consistency: Engineered prompts ensure every artifact follows the same structure - Speed: Go from idea to detailed PRD in 5 minutes instead of 2 hours - Quality: Forces PMs to think through context (market, risks, constraints) upfront - Learning: Reverse-engineer the prompts to understand PM best practices - Demonstration: Shows both AI fluency AND PM rigor (the outputs are opinionated, not generic)

Who Needs This: - PMs who want AI-assisted drafting without losing rigor - Interviewers evaluating your AI + PM thinking - Founders building products quickly with structured thinking - You, demonstrating prompt engineering depth

How

Architecture:

User Input Form (product brief)
    ↓
FastAPI Route receives data
    ↓
Calls OpenAI with engineered prompt (includes examples + constraints)
    ↓
Streams response to frontend (real-time generation UI)
    ↓
Parses output into structured markdown
    ↓
Renders in editor + export options
    ↓
Save to SQLite history

Key Design Decisions:

  1. Engineered Prompts: Each mode has a detailed system prompt that:
  2. Specifies output format (markdown with heading structure)
  3. Provides examples of good outputs
  4. Lists what NOT to do (avoid generic fluff, be specific)

  5. Includes context (who the user is, why structure matters)

  6. Streaming UI: Show generation in real-time so users feel the AI is “thinking” (better UX than waiting for completion)

  7. Editable Output: Generated content is starting point, not final answer. Users edit inline before export.

  8. Export Options: Markdown for version control, PDF for sharing with executives.

  9. History: SQLite stores previous generations so users can compare versions and iterate.

Build Path:

  • Phase 1 (Week 1-2): PRD Generator only
  • Form input → OpenAI call → markdown output → export

  • Test prompt quality with 5+ examples

  • Phase 2 (Week 3): Roadmap Builder + Story Writer

  • Reuse prompt patterns from PRD generator

  • Add dependency visualization (roadmap view)

  • Phase 3 (Week 4): Polish

  • Inline editor
  • History/versioning
  • Export to PDF
  • Dark mode

Technical Stack

  • Backend: FastAPI (async, streaming support)
  • AI: OpenAI API (gpt-4o-mini for cost efficiency)
  • Frontend: HTMX + Tailwind (real-time updates without page reloads)
  • Storage: SQLite (save history and past generations)
  • Export: Python markdown + weasyprint (PDF generation)
  • Editor: Monaco Editor (code/markdown editing experience)

Why Build This Project

  1. Demonstrates Prompt Engineering — Anyone can call ChatGPT. Few can engineer prompts that consistently produce high-quality output.

  2. Shows PM + Technical Thinking — Not just “AI is cool.” You’re thinking about structure, completeness, edge cases—pure PM thinking implemented through AI.

  3. Immediately Useful — You’ll use this for your own product planning. Other PMs will want it.

  4. Portfolio Differentiator — Most portfolios show what you’ve shipped. This shows how you think + accelerate thinking with AI.

Next Steps

See strategy/04_AI_PM_TOOLKIT.md for detailed technical specification and v2 roadmap.

Expected Timeline: 3-4 weeks for full build Complexity: Medium (prompt engineering + FastAPI + export logic) Impact: High (most reusable PM tool)

Back to Projects