Exam Topic Finder: Local AI for Smarter Exam Prep

One-liner: An open-source, fully local RAG app that compares your notes with previous year questions (PYQs) and tells you which topics to study first.

Problem

Students have notes in one place and PYQs in another. Figuring out which topics repeat in exams, and where they are covered in the notes, is slow and manual. Beginners often don’t know where to start. Cloud AI tools can help, but they need internet, may be paid or limited, and require uploading personal notes to someone else’s server.

Solution

Exam Topic Finder lets a student upload their notes (PDF) and up to 5 PYQ sets (PDF). The app then:

Extracts the text and splits the notes into chunks
Converts chunks and questions into embeddings using nomic-embed-text
Retrieves the notes sections most similar to each PYQ question (semantic search, not keyword match)
Sends the retrieved context to Llama 3.2 to identify and explain important topics
Output
Important Topics: what to study
Important parts of notes: where to study it
Question-wise mapping: why it matters, with each PYQ linked to its topic and notes section

This makes results explainable instead of a black-box AI answer.

Tech Stack
Ollama for running models locally
Llama 3.2 for topic identification and explanation
nomic-embed-text for embeddings and semantic matching
Node.js backend, simple single-page frontend (HTML/CSS/JS)
Why Open-Source and Local
Private: notes never leave the student’s laptop
Free: no API keys or usage costs
Offline: works without internet after models are downloaded
Open: anyone can fork, improve, and self-host it
Limitations and Future Work
Llama 3.2 (3B) is small, so ranking quality can vary
Scanned PDFs need OCR support (planned)
Possible next steps: topic frequency charts, support for more file types, saving past analyses