Curricula

## Inspiration

I wanted to build Curricula because I believe AI can be an incredibly powerful learning tool. It can give students the freedom to ask questions, revisit difficult concepts, and learn at their own pace without feeling rushed or embarrassed.

However, a general-purpose chatbot can easily stray away from what a student is actually learning in class. Its explanations, terminology, and practice questions may not match the curriculum, the teacher’s methods, or the current lesson. I saw an opportunity to combine the flexibility of AI tutoring with the structure and relevance of a real classroom.

Curricula grounds its AI tutor in approved curriculum and the teacher’s unique, up-to-date class materials—including lessons, assignments, schedules, and released assessments. This allows students to receive personalized support while remaining aligned with their course.

The same foundation could eventually help teachers understand how their students are learning. Future versions could provide privacy-conscious analytics about common misconceptions, difficult topics, and the types of explanations that help individual students. Teachers could use those insights to adapt lessons and tailor their teaching strategies while remaining in control of the classroom.

## What it does

Curricula gives students an AI tutor grounded in what their class is actually learning.

The OpenAI Build Week demonstration focuses on one complete Grade 9 Algebra 1 classroom. It combines an approved base curriculum with current materials supplied by the teacher.

A student can:

• Ask for help with a question they got wrong on a released class quiz.
• Receive a worked explanation grounded in the actual quiz and curriculum.
• Request an interactive graph or visualization of the concept.
• Start a new conversation without losing their previous chat.
• Prepare for next week’s lessons using the teacher’s dated course schedule.
• Generate and complete a class-aligned interactive quiz.
• Receive immediate feedback, a final score, answer review, and the ability to retry.

The teacher workspace allows classroom materials to be uploaded, downloaded, and completely removed. Uploaded materials become part of the context available to the tutor, allowing teachers to keep student support aligned with their current classroom.

## How I built it

Curricula is a Next.js and TypeScript application deployed on Vercel. Supabase provides authentication, PostgreSQL data storage, and private storage for teacher-uploaded files.

The live tutor uses GPT-5.6 through OpenAI’s Responses API. The Conversations API preserves chat state, while File Search grounds responses in two separate sources:

1. An approved Algebra 1 base-curriculum vector store.
2. A classroom vector store containing current teacher- provided materials.

Keeping these sources separate allows the base curriculum to remain stable while teachers add class-specific lessons, schedules, assignments, and released assessments.

I also created hosted Quiz and Visualization Skills. The Quiz Skill produces validated interactive questions with clickable answers, explanations, progress, scoring, review, and retry. The Visualization Skill creates native graphs and other interactive learning visuals inside the conversation.

Curricula existed as an early product foundation before Build Week. During the event, I meaningfully extended it with the teacher-governed classroom workflow, separate retrieval layers, licensed Algebra 1 curriculum corpus, GPT-5.6 tutor, interactive Skills, date-aware lesson preparation, shared-demo security, testing, deployment, and final judge experience.

## How I used Codex and GPT-5.6

I developed the Build Week extension through an agent-led collaboration with Codex using GPT-5.6 Sol with High reasoning.

I supplied the product vision, classroom problem, feature requirements, source materials, ongoing feedback, and key product, design, security, and engineering decisions. Under my direction, Codex generated and refined most of the post-baseline implementation.

Codex contributed to:

• Product architecture and implementation planning.
• Frontend components and the student and teacher experiences.
• Next.js API routes and streamed chat behavior.
• Supabase schemas, migrations, authorization, and private storage.
• OpenAI Responses, Conversations, and File Search integrations.
• Curriculum and teacher-material ingestion workflows.
• Interactive Quiz and Visualization Skills.
• Security controls, automated checks, deployment, and documentation.

GPT-5.6 contributed in two distinct ways. GPT-5.6 Sol with High reasoning powered my development collaboration with Codex. The deployed Curricula tutor also runs GPT-5.6 Sol through the Responses API with Medium reasoning, File Search, and the hosted education Skills.

## Challenges

One of the largest challenges was making the tutor flexible without allowing it to lose connection with the classroom. I needed the system to combine a broad curriculum foundation with current teacher material while keeping the two sources separate and manageable.

Another challenge was safely streaming interactive quizzes and visualizations. Student-facing explanations needed to appear naturally, while incomplete structured data had to remain hidden until it was complete and validated.

The shared judging environment also required careful security work. The demonstration uses one restricted account, private file storage, server-only access to secrets, bounded uploads, request quotas, and server- authorized data access. Personal or confidential student information is not required for the demo.

Finally, because Curricula existed before Build Week, I needed to clearly distinguish the original foundation from the meaningful functionality added during the competition. Dated commits, validation records, and Codex session evidence document that boundary.

## Accomplishments that I’m proud of

I am especially proud that Curricula is a working product experience rather than only a technical proof of concept.

The demonstration connects teacher materials, curriculum retrieval, GPT-5.6 explanations, interactive visualizations, lesson preparation, and quizzes into one coherent learning flow. A student can move from understanding a previous mistake to visualizing the concept and then practicing what comes next.

I am also proud of the teacher-governed approach. AI remains flexible and personalized, but the teacher controls the classroom context available to it.

## What I learned

I learned that a strong educational AI experience requires much more than connecting a chatbot to a model.

The quality of the experience depends on trustworthy context, clear boundaries, thoughtful retrieval, teacher control, secure data handling, and interactive ways for students to apply what they learn.

I also learned how effective Codex can be as an engineering collaborator when it is given clear requirements, continuous feedback, and explicit product boundaries. It accelerated implementation across the full stack while allowing me to remain responsible for the direction and final decisions.

## What’s next

The current demonstration intentionally focuses on one complete Algebra 1 classroom. The next step is expanding Curricula into a multi-class platform with separate teacher and student accounts, classroom enrollment, per- class retrieval, teacher-editable learning priorities, and structured assignments and assessments.

Over time, Curricula could also provide teachers with privacy-conscious analytics and feedback. It could identify common misconceptions, concepts students repeatedly struggle with, and which explanations or learning formats are most effective for each student.

That would allow Curricula to support both sides of the classroom: helping students learn at their own pace while helping teachers adapt their instruction to the needs of their students.

Built With

• codex
• gpt-5.6
• katex
• next.js
• postgresql
• react
• supabase
• tailwind
• typescript
• vercel