I write the requirements, direct the AI, and test everything it produces. AI output fails in ways that look correct until somebody checks, and checking is the part I'm best at.
Plot Armor: StoryForge
Personalized children's book generator, live with checkout.
- My role
- Product design, AI-assisted build, pricing and packaging.
- How I tested it
- Bought it myself as a customer and ran the full purchase flow end to end before it went on sale.
Generative AIPayhip checkoutE2E testing
EVOLVE – Productivity Forge
AI-personalized productivity web app.
- My role
- Built in React via Lovable with GitHub sync. Designed a freemium model: free users get 3 AI-generated plans a month.
- What I learned
- Feature creep is a quality problem. I'm now cutting it back to the core.
ReactLovableLLM features
Open the app →
Video production application
Turns written scripts into finished, consistently branded short videos.
- Stack
- Python, SQLite, FFmpeg, local AI models.
- Used for
- My promotional short videos and a weekly NFL video series on TikTok, Reels and Shorts.
PythonFFmpegLocal AI
Stained glass pattern generator
Text prompt in, buildable glass pattern out.
- The defect
- The language model produced coordinates that looked right but didn't close into cuttable shapes.
- The fix
- Re-architected the pipeline: an image model makes the picture, then deterministic code traces the contours. Each tool does what it's reliable at.
LLM failure analysisPython / SVG
Gridiron Legends
Browser football RPG with creature teams.
- My role
- Wrote the full game-design bible and playable prototypes. Playtested, found the core loop felt like a generic mobile game, and re-scoped it around turn-based play with friends.
Game designPlaytesting
Two published novels
A comic fantasy series on Amazon KDP and Kindle Unlimited.
- The pipeline
- Series bible, AI drafting, multi-pass editing rubrics and beta-reader feedback rounds. I directed and edited every pass.
AI content opsEditorial QA