The inspiration for this project came from a simple frustration: most AI tools respond the same way no matter the situation. Whether you’re making a serious decision, planning strategy, or asking for honest feedback, the AI’s reasoning style rarely changes. We wanted to explore what would happen if users could control how an AI thinks, not just what they ask.
The core idea was to treat AI reasoning as a configurable system. Instead of one generic chatbot, we designed multiple thinking modes such as Decision, Strategy, Think, Truth, and Reflect. Each mode represents a different way of processing information and produces fundamentally different outputs for the same input. This shifts the AI from a conversational tool into a structured reasoning engine.
We built the project using the Gemini API for language generation and a custom backend that dynamically assembles context for every request. Each request combines a fixed base prompt, a mode-specific prompt, and a curated dataset of reasoning frameworks before being sent to the model. On the frontend, we designed a retro 1980s inspired interface to reinforce the idea that the user is interacting with a purposeful system, not a casual chat app.
Throughout the project, we learned how much control prompt structure and context design give over large language models. Small changes in constraints and framing had a major impact on output quality and consistency. We also learned the importance of clear UX in explaining complex system behavior to users.
The main challenges were balancing flexibility with predictability and avoiding vague or generic responses. Designing distinct modes that felt meaningfully different without overcomplicating the system took several iterations. Another challenge was ensuring the AI stayed within the provided context and didn’t drift into unsupported assumptions.
Overall, the project demonstrates that with thoughtful system design, a single language model can behave like multiple specialized reasoning tools, giving users clarity, control, and intentional interaction.
Built With
- geminiapis
- html5
- typescript
Log in or sign up for Devpost to join the conversation.