ReasonMap

Inspiration

People are often told to “think critically,” but rarely given a calm, usable process for doing it. Most AI tools move quickly toward answers. I wanted to build something that does the opposite: help a person slow down, examine their reasoning, and make the discovery for themselves.

What it does

ReasonMap is a guided reflection tool for decisions, uncertainties, and questions without obvious answers. It leads the user through seven stages:

1. Frame the question
2. Identify what matters
3. Examine the evidence
4. Surface assumptions
5. Consider perspectives
6. Check for reasoning gaps
7. Reflect and summarize

The user stays in control throughout. They can choose Guide Me, Challenge Me, Let Me Think, or Help Me Organize My Thoughts. They can also ask for a hint or move on at any time.

As the conversation develops, ReasonMap builds a visible map of what the user has actually shared: priorities, evidence, assumptions, perspectives, and remaining gaps. It reveals the map gradually so that users are not overwhelmed or made defensive. The final screen reflects the user’s reasoning without turning it into a recommendation. The decision remains theirs.

How I built it

ReasonMap is a responsive TypeScript and React application built with Codex and GPT-5.6. The current Build Week version uses Next.js-compatible components, Vinext/Vite, and Cloudflare-based hosting.

I used Codex throughout the product-development process: turning a detailed educational design philosophy into an interface, building the staged session flow, implementing mode changes and the live reasoning map, refining prompts, and diagnosing interaction problems. GPT-5.6 helped preserve the important distinctions in the design—for example, asking one useful question at a time, separating guidance from decision-making, and keeping assumptions internal until the learner is ready to examine them.

Challenges

The hardest part was restraint. A reasoning assistant can easily become pedantic, repetitive, or too eager to summarize. I had to design the interaction so that it asks enough to support deeper thinking without taking ownership of the decision. Another challenge was making progress visible without presenting a long, intimidating list of inferred assumptions.

Accomplishments

I am proud that the interface makes sophisticated reasoning practices feel quiet and approachable. The mode controls give the user real agency. The reasoning map records only confirmed ideas, and the final reflection explicitly avoids issuing a verdict.

What I learned

The most useful AI interaction is not always an answer. Sometimes it is a well-timed question, a pause, or a structure that helps someone notice the next step for themselves. Building ReasonMap also showed me how much product quality depends on small conversational details: when to ask, when to stop, and when to let silence do the teaching.

What’s next

The next version will connect the interface to live GPT-5.6 guidance, strengthen the internal reasoning-state model, add optional user-selected labels, and support reusable educator-designed reasoning activities while keeping the clean, private-by-design experience.

Built With

• cloudflare
• codex
• gpt-5.6
• next.js
• react
• typescript
• vinext
• vite