Academic Assistant Pro– AI Tutor with Retrieval-Augmented Generation (RAG)

Abstract

The rapid advancement of artificial intelligence and natural language processing has opened new avenues for enhancing educational experiences. Traditional digital learning tools, while effective in delivering static content, lack the interactive, adaptive, and per sonalized nature of human tutoring. The emergence of Large Language Models (LLMs) has begun to bridge this gap, but their well-documented tendency to produce hallucinated or unverifiable information has limited their adoption in academic settings where factual accuracy is non-negotiable. This project presents Academic Assistant Pro, an AI-powered tutoring application that leverages Retrieval-Augmented Generation (RAG) to provide con textually accurate, document-specific academic assistance. Unlike conventional chatbots that rely solely on pre-trained knowledge, Academic Assistant Pro ingests user-uploaded PDF documents, chunks and embeds them into a vector database, and retrieves relevant passages to ground the large language model’s responses in verified source material. The application is built using Python with PySide6 for a cross-platform graphical user interface, LangChain for the RAG pipeline orchestration, Ollama for local large language model inference (using Qwen2.5:3b), and Chroma as the persistent vector store with nomic embed-text embeddings. The system supports multiple teaching modes– Beginner, In termediate, and Exam– tailoring response style and depth to the learner’s proficiency level. Additional features include streaming real-time responses, automated quiz generation with scoring, interactive flashcard creation, structured document summarization (with topic names and bullet points), conversation memory with history persistence, drag-and-drop PDF upload, dark/light theming, and chat export functionality. All processing occurs locally on the user’s machine, ensuring data privacy and offline capability– a critical advantage for institutions and individuals working with confidential study materials, proprietary research, or copyrighted course resources. The multi-threaded architecture, using PySide6’s QThread and QMutex primitives, ensures the graphical in terface remains responsive during computationally intensive operations such as document embedding, retrieval, and language model inference. Comprehensive testing including unit tests, integration tests, performance benchmarking, and user acceptance testing with 15 student participants demonstrated that the system achieves 92% factual accuracy on document-grounded queries, with average response times of 5–10 seconds for typical ques tions on consumer-grade hardware. Academic Assistant Pro represents a practical, privacy preserving, and pedagogically informed tool for self-directed academic learning and serves as a reference architecture for future local-first AI applications in education.