Shelly

Lift every student with affordable, AI-powered test prep that breaks barriers and builds futures.

Comment

Shelly - SHSAT Practice Platform

Inspiration

I was inspired to create this tool when I learned that the format of the Specialized High School Admission Test (SHSAT) is changing. The SHSAT is the sole factor in admissions to New York City's best public high schools. Naval Ravikant (Stuyvesant class of 1991) says of the specialized high schools: "You go from being blue collar to white collar in one move."

Historically, Black and Hispanic students have not been adequately prepared for this exam. You can analyze the disparities using this data visualization tool I built at https://shsat.nyc/. In 2025, the SHSAT is going digital with new technology-enhanced items (TEI). In 2026, it's becoming adaptive like the GMAT. I worry that these changes to the exam will only exacerbate racial disparities in outcomes.

I'm creating a free tool for students to practice the new format because only two practice exams have been officially released by Pearson Education, the exam's administrator. Two practice exams is not enough for students to successfully prepare for this exam. With advances in AI, I can develop a better product faster than Pearson.

What It Does

Shelly is a clone of the new digital ADAM testing platform made by Pearson Education. It is populated with AI-generated questions that mirror the exam. Students can familiarize themselves with new digital test-taking tools like highlighting, bookmarking, and answer elimination. The new technology-enhanced items like drag-and-drop are also currently supported in this prototype. One of my hypotheses is that students will be more motivated to study for exams by reading personalized passages catered to their interests. If they like basketball, the reading comprehension article could be about the history of the New York Knicks.

You can now analyze the exact amount of time a student spends on each question. With 114 questions and only 180 minutes, every second counts.

The platform includes an admin portal for teachers to analyze student performance.

How We Built It

Development began with establishing a robust foundation using Vue 3 with TypeScript, Vite, Tailwind CSS, and Pinia. The initial prompts focused on creating a question flow interface. Key features were iteratively added, including text highlighting capabilities (red/blue colors), answer elimination mode, bookmarking system, and comprehensive timing mechanisms for each question.

The project evolved to include advanced analytics with mock data visualization using custom SVG charts, a detailed review system showing correct/incorrect answers with explanations, and a teacher dashboard for class performance analysis. Throughout development, the codebase maintained clean architecture with modular components, proper TypeScript typing, and responsive design principles. The platform now offers a complete educational ecosystem with student practice modes, detailed performance analytics, and teacher insights.

Challenges We Ran Into

The biggest challenge was using AI to deconstruct the exam. The new exam is digital and only one question appears at a time. I had to manually tag the data. Now that I'm writing this, I realize I could have just written a scraper with AI to do this. Oops! I think the manual process helped me internalize the nuances of the new exam, though.

The new TEI problem types add significant additional complexity. I was only able to implement the drag-and-drop feature for this prototype, but analysis of the two official practice exams shows that this is the most common new TEI.

Accomplishments We're Proud Of

This prototype was used in my first sales pitch to a test prep center. I'm still currently in talks for a formal partnership, but I'll be using it for more outreach with other test prep centers. I'm dedicated to making this tool accessible to students completely free. I'll be charging test prep centers for access to the teacher portal. I'm also on the Junior Board of the Stuyvesant Alumni Association and working on a partnership with their nonprofit StuyPrep, which helps underserved communities prepare for this exam.

What We Learned

In deconstructing the exam, I learned that AI is better at parsing rich text from images than from PDFs.

I learned that building the frontend without a backend is the easiest way to move as quickly as possible with prototyping in Bolt. It reduces technical complexity. I established comprehensive JSON schemas to capture the data needed on the frontend to drive backend development.

I learned that Bolt is way better than Replit. I would explicitly tell Replit to use a particular framework or follow specific instructions, and it would straight up ignore my directions.

I relied heavily on Claude to craft prompts for Bolt.

What's Next for Shelly

The immediate goal for Shelly is to make it production-ready. I need to add support for user authentication and integrate this frontend built with Bolt with the backend deployment infrastructure I have set up. Ultimately, I need to get user feedback as soon as possible to drive the product forward. I hope for Shelly to generate revenue this summer.

Built With

Share this project:

Updates