What I'm Building

A consumer app where people train AI models to understand their unique taste and judgment in specific domains through fun, game-like interactions. Instead of optimizing for what's objectively good or bad, the app learns what you think is good. It builds a living model of your taste that evolves over time. That model can be exported so any AI instantly understands how you think.

I'm exploring a consumer app where people can train AI models to understand how they think in specific areas instead of repeatedly prompting them. Users pick a domain like writing, design, shopping, or create their own custom module, then teach the model through fun, game-like interactions. Over time, the app builds a living model of their taste and judgment that evolves as they do. The profile belongs to the user, not the AI company, and can be exported in an AI-native format so any model can instantly understand them. Unlike AI training gyms that optimize models around what's objectively good or bad, this optimizes models around each individual's unique taste and judgment.

I'm exploring a consumer product idea and want to think through it from first principles. The core idea is that everyone develops a unique sense of taste and judgment over time, but today's AI never really learns it. Every time we use ChatGPT, Claude, or another model, we end up saying things like "make it sound more like me," "I wouldn't choose that," or "that's not how I think." Those preferences live inside individual conversations and stay locked inside each AI. I'm wondering if there should be a product whose only job is helping people build portable taste profiles that belong to them, not to a specific AI. Instead of building one giant personality profile, users pick one area they want an AI to understand better. That could be: • Writing style • Design taste • Product judgment • Hiring decisions • Shopping preferences • Travel planning • Communication style • Decision making Users can also create their own modules around things they do all the time. Maybe it's firing email style, performance reviews, investment decisions, code reviews, sales messaging, customer support, or any workflow where judgment matters. Each module feels like a game instead of work. Users make lots of small decisions over time. They choose between options, critique AI outputs, rank ideas, edit responses, or explain why one choice is better than another. The goal isn't to learn isolated preferences. It's to understand how someone thinks, what tradeoffs they make, and what "good" looks like to them. The profile is never finished. People's taste changes as they gain experience, switch careers, or simply change their mind. The profile should evolve with them. At any point, users can export their profile in a format that any AI can understand. Instead of teaching every new model from scratch, they import their profile and the AI starts with a much better understanding of how they think in that domain. The hardest part isn't the export format. It's designing interactions that people genuinely enjoy while collecting enough signal to model real judgment. I also don't want these profiles to become a list of likes and dislikes. They should capture context, tradeoffs, exceptions, and how someone's judgment changes depending on the situation.