ClearSheet

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for ClearSheet

Inspiration

ClearSheet began with a real problem at home. My son is in Year 3 and has suspected learning differences, possibly including dyslexia. Some homework sheets were unnecessarily difficult for him to interpret—not because of the academic content, but because the instructions, presentation and visual structure created avoidable barriers.

I understand why this happens. Teachers have limited preparation time, and their lessons naturally take priority over redesigning routine homework resources. I researched inclusive learning design and built a repeatable checklist and design profile. Before Build Week, I used those materials manually in ChatGPT: first improving the language, then approving it, then creating a visual worksheet. The results could be excellent, but the workflow required detailed prompting and several stages.

Before Build Week, ClearSheet did not exist. Build Week gave me the opportunity to use Codex to turn that personal workflow into a purpose-built application that teachers could use quickly.

What it does

ClearSheet transforms an existing PDF, image or Word worksheet into a clearer, more inclusive resource while preserving its learning objective and academic challenge.

The teacher:

1. uploads an existing worksheet;
2. checks a short explanation of the proposed changes;
3. edits, requests changes or approves the redesign;
4. reviews the actual printable worksheet; and
5. downloads a PDF or editable Word version.

The aim is not to make the work easier. It is to remove unnecessary access barriers while keeping the teacher in control. Clearer instructions, stronger visual hierarchy, purposeful support and suitable response space can help learners with additional needs while also benefiting the wider class.

How I built it

I used Codex with GPT-5.6 to build ClearSheet from scratch during Build Week. Codex inspected my research, checklist, design profile, source worksheets and previous redesign examples, then helped turn them into a complete application.

The product went through around five iterations. Early versions exposed too much analysis, required too many teacher decisions, or could not consistently match the quality of the earlier manual workflow. Working with Codex, I simplified the journey, strengthened source-content preservation, added teacher approval gates, improved one-page layout logic, built PDF and editable Word exports, added deterministic quality checks and created a protected judge deployment.

ClearSheet itself uses GPT-5.6 to understand the uploaded worksheet, identify access barriers and produce the inclusive content and page plan. Once the teacher approves that content, GPT Image 2 creates the visual worksheet preview. GPT-5.6 then checks the finished result against the locked content. The app keeps the OpenAI API key on the server.

The application is built with Python and FastAPI, with a focused HTML/CSS/JavaScript interface. It is deployed on Render, backed by automated tests, and includes sample worksheets and full setup instructions in the public repository.

Challenges

The hardest challenge was balancing inclusion with academic integrity. ClearSheet must clarify the route through a task without lowering expectations, inventing content or completing the pupil's work.

Another challenge was reliability. Uploaded worksheets can be PDFs, Word files or images, and their structure varies widely. The app needs to preserve important diagrams, questions, vocabulary and answer requirements while still creating a calmer design.

Visual generation also needed safeguards. The approved wording is locked before preview generation, exports remain unavailable until the teacher approves the result, and deterministic checks provide a safe fallback when a generated visual does not meet the content contract.

Accomplishments

I am most proud that ClearSheet now reproduces the core quality and control of my original manual process in a workflow designed for busy teachers. A teacher can go from an existing worksheet to a reviewed, printable result without learning prompt engineering.

I am also proud that the live judge deployment runs the real GPT-5.6 and GPT Image 2 pipeline, rather than a simulated demonstration.

What I learned

Inclusive design is not about reducing challenge. Small changes to language, hierarchy, visual support and response space can change whether a pupil can access the intended learning at all.

I also learned how important iteration is when building with generative models. Clear prompts matter, but so do typed content contracts, approval gates, validation and deterministic fallbacks.

What's next

ClearSheet is a working Build Week prototype, not a finished school platform. A production version would need school-grade authentication, tenant isolation, secure persistent storage, retention controls and further testing with teachers and learners.

The next step is to test the workflow with educators, refine subject-specific layouts and explore safe school deployment. The goal remains simple: help teachers create clearer, more inclusive worksheets without adding to their workload.

Built With

• gpt
• image
• openai