PasoClaro AI

Inspiration

Many learners do not struggle because they lack ability, but because explanations skip the exact step where their understanding breaks. A correct solution can still be difficult to follow when it assumes that the learner already understands an intermediate operation, rule, or inference.

PasoClaro AI was inspired by the need for educational support that slows down, identifies the first point of confusion, and rebuilds understanding without overwhelming the learner.

What it does

PasoClaro AI is a step-by-step learning assistant. It receives a problem, the learner's attempted solution, and an explanation of what they find confusing.

The system is designed to:

• identify the first unsupported or misunderstood step;
• explain only that step in clear language;
• avoid unnecessary transformations or hidden assumptions;
• ask a short question to verify understanding;
• continue only when the learner is ready.

Instead of immediately giving the complete answer, PasoClaro AI aims to preserve learner agency and reduce cognitive overload.

How we built it

PasoClaro AI is currently being developed during OpenAI Build Week as a focused web prototype.

I am using Codex to assist with project structure, implementation, debugging, testing, and documentation. GPT-5.6 will analyze the learner's reasoning and generate adaptive explanations one step at a time.

This section will be updated before submission with the final technical architecture and the tools actually used.

Challenges we ran into

The main challenge is distinguishing between an incorrect answer and the precise point where understanding failed.

Another challenge is preventing the system from explaining too much at once. PasoClaro AI must remain useful without overwhelming the learner or simply replacing the learner's own reasoning.

This section will be updated as development and testing continue.

Accomplishments that we're proud of

The project has been clearly defined around a specific educational problem: explanations that are technically correct but cognitively inaccessible.

The current scope prioritizes one functional learning workflow instead of a large platform with incomplete features.

The final accomplishments will be documented after the working prototype has been tested.

What we learned

The initial design process has reinforced an important principle: educational assistance should not only produce correct answers. It should also detect assumptions, missing steps, and the learner's actual point of confusion.

This section will be expanded with lessons from implementation and user testing.

What's next for PasoClaro AI

The immediate goal is to complete and test the working prototype for OpenAI Build Week.

Future development could include:

• support for additional subjects;
• accessibility options;
• learner-controlled explanation depth;
• multilingual guidance;
• progress tracking;
• teacher and tutor feedback tools.

Built With

• api
• codex
• gpt-5.6
• openai
• python
• streamlit