Peel

GitHub: https://github.com/Team-VSM/SuvidhaHackathon Live Preview: https://suvidhahackathon.backslash2223.workers.dev/

Inspiration

Peel began with a problem that is easy to mistake for laziness: sometimes, the hardest part of a worksheet is simply starting it.

One of our team members, Michael, has ADHD, so this challenge was personal to our group. A worksheet containing twenty manageable questions can still appear as one enormous, undefined task. Before the student answers anything, they must process the layout, estimate the time required, decide where to begin, and keep track of their progress. That initial mental load can make the entire page feel inaccessible.

We wanted to build an accessibility tool that reduces this barrier without completing the work for the student. Peel is therefore not an answer generator or cheating tool—it is an initiation accommodation designed to help students make the first mark.

What it does

Peel transforms a worksheet into a focused, one-question-at-a-time experience.

A student can photograph a worksheet, upload an image, paste a screenshot, or enter its text. Peel uses AI to extract the questions and estimate the effort each one may require. Before showing the work, it converts the unknown page into three understandable numbers:

• The total number of questions
• The estimated completion time
• The number of questions likely to require more effort

Instead of committing to an entire worksheet, the student commits to completing three questions and then deciding whether to continue. Peel begins with the easiest question, displays only one question at a time, and lets the student temporarily park anything they are not ready to attempt.

Other accessibility features include:

• A four-level Unstick hint system that explains without giving away answers
• Read-aloud support
• Adjustable themes and text sizes
• A scratchpad and calculator
• Quiet-company mode
• A pace bar that communicates time without creating a failure state
• Exact session restoration, including unfinished writing
• A locally stored teacher or parent report showing where the student slowed down, deferred a question, or returned to it

Peel deliberately avoids points, streaks, feeds, accounts, and notifications. We did not want an accessibility tool competing for the same attention it is supposed to protect.

How we built it

We built Peel as a responsive React 19 application using TypeScript, Vite, and Tailwind CSS.

The frontend handles worksheet uploads, image compression, question presentation, accessibility controls, progress tracking, and local session storage. Mobile camera images are resized and compressed in the browser before processing, making uploads faster while keeping the text readable.

A Cloudflare Worker provides two primary AI endpoints:

1. /api/extract-questions analyzes a worksheet image or its text, extracts individual questions, and estimates their effort.
2. /api/unstick generates a progressive hint ladder while being instructed not to reveal the answer.

We use Featherless AI with the Qwen3.8-27B model for worksheet understanding. The OpenAI-compatible SDK connects the application to the model API.

Peel also records privacy-preserving measurements such as:

• Time to first mark
• The point where a session was abandoned
• How often deferring a question led to continuing rather than quitting
• An optional dread rating before and after a session

These measurements remain in the browser through localStorage and can only leave the device when the user explicitly exports them as JSON.

Challenges we faced

Extracting messy worksheets

Worksheets are rarely formatted consistently. They can contain numbered questions, instructions, diagrams, multiple sections, handwriting, or low-quality camera images. We had to make the extraction process tolerant of many layouts while still returning structured questions in a predictable format.

Providing help without providing answers

It would have been much easier to build another AI homework solver. Instead, we needed Peel to support learning without replacing it. Designing the Unstick system required carefully separating rewording, answer structure, first steps, and decomposition from simply revealing the solution.

Designing for initiation rather than productivity

Many productivity applications assume that users have already begun. Peel focuses on the moment before that—the point where the page still feels too large to approach. That forced us to reconsider familiar design patterns. Streaks, countdown timers, large task lists, and aggressive reminders can create more pressure, so we intentionally left them out.

Being honest about the evidence

Chunking is frequently recommended for students with ADHD, but the direct research supporting it is limited. Existing research often measures accuracy or time on task after a student has already begun. Peel instead focuses on time-to-first-mark, because initiation is the specific barrier we designed it to address.

We learned that building responsibly means clearly separating what a product measures from what it has already proven.

Preserving privacy

Worksheets, notes, timing data, and learning difficulties can be sensitive. We designed session data and metrics to remain local by default, without requiring accounts or building an attention-driven platform around the student.

What we learned

We learned that accessibility is not always about simplifying the educational material. Sometimes the material is already understandable, but its presentation creates the barrier.

We also learned that small interface decisions can carry significant psychological weight. Showing one question instead of twenty, providing a non-punitive “Not now” option, or replacing an uncertain workload with a concrete estimate can change whether a task feels approachable.

Technically, we gained experience with multimodal AI, prompt constraints, image preprocessing, Cloudflare Workers, responsive accessibility design, local-first data storage, and instrumentation for measuring user behavior without invasive tracking.

Most importantly, Michael's perspective helped us remain focused on the real experience behind the project. We were not trying to “fix” ADHD. We were building a tool that respects how different students begin difficult tasks.

What's next

Our next step is to test Peel with more students and measure whether it meaningfully reduces time-to-first-mark. We also want to improve extraction for diagrams and unusual worksheet layouts, expand read-aloud controls, make effort estimates more accurate, and give educators clearer—but still privacy-respecting—ways to understand where students become stuck.

Peel's goal is simple: make the first question feel possible.

Built With

• featherless
• react
• tailwind
• typescript
• vite