LectureMind – Fully Local AI Study Assistant Powered by RAG

A complete Python project that transforms lecture PDFs into interactive study tools, running entirely locally with no external API calls and no data leaving the device. Key features include:

PDF text extraction and cleaning (PyMuPDF) with deduplication

Automatic summarization using BART with a map-reduce approach

Key topic extraction via agglomerative clustering of sentence embeddings

A 4-stage Retrieval-Augmented Generation (RAG) Q&A pipeline: FAISS bi-encoder retrieval → cross-encoder re-ranking → cosine-similarity deduplication → a single call to a local LLM via Ollama

Automatic multiple-choice question (MCQ) generation with embedding-validated distractors and duplicate-question filtering

A full Streamlit front-end tying everything together

و

The project reflects deliberate engineering decisions for example, using page-based chunking for slide decks (since each slide represents one concept) instead of word-count chunking, and avoiding clustering in MCQ generation since the cluster-count parameter is known to be unstable.