The problem

Finding the right passage in a long set of notes can be difficult when the question uses different words from the source. StudyBridge helps a learner return to the original text instead of relying on an unsupported generated answer.

What it does

Paste notes, prepare them, and ask a question. StudyBridge ranks passages by semantic similarity and displays the source passages. It uses the Xenova/all-MiniLM-L6-v2 sentence embedding model in the browser through Transformers.js. The model and library are downloaded on first use; the notes and question are processed locally.

How it was built and tested

The interface uses HTML, CSS and JavaScript. Embeddings are compared with cosine similarity. A photosynthesis example was tested with a question about oxygen and returned the relevant paragraph. Editing notes disables search until the notes are prepared again.

Limitations

This is a retrieval tool, not a factual answer generator. A high similarity score does not establish truth, and users should verify results in context. It is English-focused and requires a first-time model download.

Credits and AI disclosure

OpenAI Codex assisted with the design and implementation. The app uses Hugging Face Transformers.js, ONNX Runtime Web and the Xenova/all-MiniLM-L6-v2 model. The project was started during the GenHacks period.

Built With

Share this project:

Updates

Submission history