

The secret to better AI prototypes: Why Tinder’s CPO starts with JSON, not design | Ravi Mehta (product advisor, previously EIR at Reforge)
205 snips Sep 29, 2025
Ravi Mehta, former Chief Product Officer at Tinder and a seasoned product advisor, shares his innovative approach to AI prototyping. He critiques the prevalent 'vibe prototyping' that yields mediocre results and advocates for using JSON data models as a more effective foundation. Ravi introduces a three-part framework for crafting stunning Midjourney prompts and emphasizes the significance of real data over Lorem Ipsum. He also reveals tricks for elevating AI-generated images, making this a treasure trove of knowledge for product creators.
AI Snips
Chapters
Transcript
Episode notes
Avoid One-Prompt Jack-Of-All Trades
- Asking one model to do UX, data, and code at once produces average results across each area.
- Separating concerns yields better-focused outputs for design, data, and engineering tasks.
Start Prototypes With JSON Not Vibe Prompts
- Generate structured JSON data first and paste it into your prototyping tool to build realistic, functional prototypes.
- Use real image URLs via tools like Unsplash MCP to avoid hallucinated media and improve stakeholder feedback.
Use LLMs To Create Rich Mock Data
- Use an LLM (e.g., Claude) to generate detailed mock JSON that includes nested fields, messages, and tags for realistic behavior.
- Paste that JSON into your prototyping tool and let the UI logic be driven by authentic data.