Cognitive Companion
An AI personality customizer built with OpenAI's API
LLMs are infinitely flexible — but most interfaces expose none of that flexibility to the end user. The result is interactions that feel generic and impersonal.
The hypothesis: if people could shape their AI's personality upfront, interactions would feel more intentional, more useful, and more engaging.
- Personality sliders — tone (formal ↔ casual), verbosity, directness
- Domain focus — users tag areas of interest to prime responses
- Emotional register — warm, neutral, or analytical response style
- System prompt builder — generates a refined system prompt from user inputs
Built with Next.js, OpenAI API (GPT-4), and a lightweight JSON schema that maps user personality inputs to system prompt segments. Personality configurations are saved per-session and exportable.
The project raised a genuinely interesting design question: how much control is too much? Users with more customization options reported higher satisfaction — but also spent more time in setup than conversation. There is a real tension between configurability and time-to-value.