AceSAT Hackathon Submission

Project Name: Beacon (Offline-first SAT coach)
Video & Working Demo (3 minutes): YouTube Video
One-Page Write-Up: Google Document
GitHub Repository: Repo

Inspiration

Private SAT tutors can cost up to $100 an hour. For many students in underserved schools, personalized support can be out of reach. Even free online tools assume that students have access to reliable internet connection.

When we prepared for the SAT, our most useful tool was a simple Excel spreadsheet to log every missed question and why. We could download and review it offline, but it only recorded mistakes. It couldn't turn these mistakes and reflections into a personalized study plan. Beacon can. Beacon acts as an adaptive SAT coach.

What it does

When connected to WiFi, Beacon reads the student's practice history, decides where improvement will matter most, and packs a training route around their study time, target score, and test date. With that, the student can practice offline.

The core loop is:

CONNECT → ANALYZE → PLAN → PACK → GO OFFLINE → PRACTICE → RETURN → LEARN → RESYNC

Only the first four phases and the last two use a brief connection. Practice and review run entirely offline.

Before answering a question, students rate their confidence. For missed questions, students name the cause and the agent writes a personalized lesson learned or reflection. These reflections become a structured Review and Progress log, which separates what a student doesn't know from what they thought they knew. Students can also select the tone of their Beacon-generated reflections to be constructive, Socratic, direct, or encouraging based on their preferred learning style.

If a student has 20 minutes to study, they can practice with a full targeted session. Even if the student only has 5 minutes to spare, Beacon can recommend the single highest-value action. Every recommendation comes with the reasoning behind it.

Beacon doesn't assume its recommendations and predictions are always right. Each decision carries a falsifiable prediction corresponding to a specific metric. On the next WiFi check-in, Beacon scores its initial prediction against recorded answers. Interventions that keep missing get demoted. The Decision Ledger records if the advice actually worked.

How we built it

Beacon is a Progressive Web App (PWA) that can be installed on a mobile device or school Chromebook. A service worker proactively caches the shell and question bank. IndexedDB stores the student's practice history. Training, review, progress, and the decision ledger continue to work with the network fully off.

The decision engine is built on TypeScript. Claude Haiku 4.5 can handle the job of rewriting computed decisions into user-facing sentences. Even without an API key, the app remains fully functional.

The practice question bank holds 161 original multiple-choice questions. The questions were carefully designed with answer choices that include tempting but incorrect options reflecting common student mistakes. This helps distinguish genuine student understanding from guessing.

Accomplishments that we're proud of

Impact: Beacon turns incorrect answers into a personalized study system by automating the metacognition process. The agent tracks errors across study sessions, identifies patterns and weaknesses, and develops a practice plan to improve their score.
Innovation: Beacon generates data-driven recommendations, tailors feedback and reflections to the student's learning style, and proactively prepares offline practice sessions. Beacon then compares its predictions to student performance, creating a feedback loop that continuously improves its recommendations.
Technical execution: This project was built with TypeScript, Next.js 15, and IndexedDB. Beacon was developed through spec-driven development with AI assistance. The conceptual framework of this project was human-designed, and the AI was used to execute human-initiated prompts. We thoroughly reviewed, edited, and tested the code to ensure its functionality. Technical and product decisions were carefully documented.
Accessibility: Beacon can be accessed and run without a consistent WiFi connection. This allows students to study anytime and anywhere. The interface also features large touch targets, different lighting modes, and reduced motion to support users.

What we learned & challenges faced

Through this project, we learned how to make a Progressive Web App (PWA) and how to design around factors like offline usage. This involved carefully writing our specifications to support that functionality through proactive data caching and a model of syncing or "docking" when a WiFi connection is available and then preparing for when there isn't one.

We faced an initial challenge in generating effective SAT-like problems since College Board's official practice material is copyrighted and already widely practiced. So, we developed our own question bank by referencing the concepts and learning goals the SAT tests and reviewing common challenges that students encounter.

What's next for Beacon

Growing our question bank to include a wider variety of questions, passages, and topics.
Supporting score predictions and recommendations on the scale of weeks or months of training, rather than session by session.

Thank you so much for reviewing our project.