PreguntaLab

Inspiration

I am a secondary-school social studies teacher, not a professional programmer. I created PreguntaLab because many of my students struggle with Colombia's Saber exams, especially with the type of reasoning required by Saber-style questions.

Students need frequent practice, but producing a good assessment is demanding. A teacher must find reliable source material, design the context, write plausible answer options, review every question, deliver the test, grade it, and explain the mistakes. Preparing one bimonthly assessment could take me a week or more, which made continuous practice almost impossible for one teacher.

When ChatGPT became publicly available in November 2022, I saw that artificial intelligence could help turn an educational idea I had carried for years into something practical. At first, I copied and pasted examples and asked AI to help me understand their structure. As the tools improved, I began experimenting with small projects and teaching myself concepts I had never used before: code, Git, repositories, databases, deployments, and testing.

AI did not instantly make me a programmer. It gave me a bridge—a way to shorten the distance between a classroom problem I understood deeply and a software solution I wanted to build.

What it does

PreguntaLab supports a real classroom assessment workflow. A teacher shares a link with students. Each student enters a name, course, an individual key, and a course access key. The platform creates an assessment session, assigns questions, records answers, calculates the score and grade, and shows which responses were correct or incorrect so the student can learn from the result.

The grade can then be sent through Google Apps Script to the teacher's Google Sheets gradebook. Teacher-facing routes support assessment creation, question management, activity monitoring, and results review.

The current submission focuses on the working formal-assessment experience. Individualized question assignment and anti-fraud controls already exist but still require refinement. A broader practice mode with more personalized guidance is part of the roadmap.

How I built it

I built PreguntaLab from the ground up through an AI-assisted, self-directed learning process. I remain responsible for the educational purpose, assessment rules, product decisions, classroom scenarios, and approval of every change. AI tools helped me translate those decisions into software, and Codex became especially valuable because it could inspect the complete repository, understand context across files, diagnose problems, and help me test changes systematically.

The application uses Next.js, React, TypeScript, Tailwind CSS, Supabase, Vercel, Google Apps Script, and Google Sheets. It contains separate student and teacher experiences, API routes, controlled access, assessment sessions, question assignment, answer validation, scoring, feedback, and grade synchronization.

PreguntaLab existed before OpenAI Build Week. During the submission period, I meaningfully extended and stabilized it with Codex. This work included improving question assignment by difficulty, diagnosing a mismatch between the questions assigned to a student session and those loaded by the real student interface, repairing session-based question loading, validating the complete assessment-to-grade flow, and documenting a safer testing process.

Challenges

My biggest challenge was not writing an isolated feature; it was making the entire classroom journey reliable. A successful build did not prove that a student could enter, receive the correct questions, submit answers, store a result, see feedback, and send a grade to Google Sheets.

One critical problem involved individualized session snapshots. The system could assign one set of questions to a session while the student page loaded another set. Codex helped trace the real flow across the frontend, API routes, Supabase, and the external grade integration. The corrected flow now carries the session identifier when retrieving questions, keeping the displayed assessment aligned with the answers validated at submission.

Security and privacy are equally important. Because this project involves students, real names, identifiers, and grades must never be used for demonstrations or QA. The judging experience will therefore use fictional credentials and isolated test data.

What I learned

I learned that educational software begins with pedagogy, but it succeeds through precise rules, careful testing, privacy protection, and an understanding of how every technical component affects the learner.

I also learned how to collaborate responsibly with Codex: begin with diagnosis, examine evidence, authorize a limited change, test the exact user journey, and document both the result and the remaining risks. This process has allowed me to build a functional product without pretending to have a traditional software-engineering background.

What's next

My goal is for PreguntaLab to become more than an online test. I want students to practice continuously, receive explanations based on their own answers, understand their weaknesses, and follow a clearer path toward improving their Saber scores.

Next steps include completing the practice mode, strengthening security and anti-fraud behavior, refining individualized question delivery, creating a permanent isolated demonstration environment, expanding the question bank, and developing more personalized AI-supported feedback.

Built With

• apps
• codex
• framer
• google
• next.js
• react
• sheets
• supabase
• tailwind
• typescript
• vercel
• zod
• zustand