Bridge

Your lecture notes, explained through what you love, with an honest note on where each comparison breaks.

Comment

Inspiration

Our lecture notes only ever contained the lecture. We remember what we love far more easily than what we are taught, so we wanted an app that connects the two: it takes our notes and explains them through our own interests.

What it does

You upload your notes as a PDF (typed, handwritten or both) or paste the text, then pick an interest, a level and a language. Bridge returns a report with the four to six key concepts of the lecture. Each one is explained first the way the notes put it and then through your interest, with a worked example and a common misconception. Every comparison also says where it stops working and gets a match score from 1 to 5, so the analogy never replaces the real subject. The report includes a concept map, a five-question quiz on the lecture, next steps, and a PDF you can download. Reports can be in English or Turkish.

How we built it

Bridge is a Next.js 14 app written in TypeScript with Tailwind. The notes and the interest are sent to OpenAI's GPT-6 Luna model. A PDF of up to 30 pages is sent as the file itself, so the model gets both the text and an image of every page, which is how it reads handwriting and diagrams. The model must answer in a fixed JSON structure; the server checks every answer with Zod and asks once more if it does not fit. The downloadable PDF is built in the browser. There is no database and no accounts. We built it as a team of two, using Claude Code as our coding assistant.

Challenges we ran into

  • Handwriting: normal PDF text extraction sees nothing in handwritten notes, so we had to send the pages to the model as images.
  • A PDF library that rejected valid small files depending on what had been uploaded before. We found it through testing and replaced the library.
  • Keeping the model's output reliable: it has to follow one exact structure every time, so we added validation and a retry.
  • Honesty: making the model admit where an analogy breaks instead of forcing a comparison.

Accomplishments that we're proud of

  • It reads typed, handwritten and mixed notes.
  • Every analogy comes with an honest note on its limits.
  • A full report costs less than half a US cent and takes about half a minute.
  • The whole flow works end to end, from upload to a finished PDF, in English and Turkish.

What we learned

We learned how to make an AI model return structured, checkable output instead of free text, and why its answers must be validated. We learned to test with our own real class notes early, because that is where the real problems appeared. We also learned to work as a team on one codebase with Git.

What's next for Bridge

  • Export the quiz and concepts to Anki flashcards.
  • A teacher mode, to prepare one lecture for a whole class with different interests.
  • A plugin for learning platforms such as Moodle or Google Classroom.
  • Handwriting support for PDFs longer than 30 pages, and faster reports.

Built With

  • apis
  • languages
  • platforms
Share this project:

Updates

Submission history