Abstract
EduMind AI is a Python-based intelligent educational tutoring web application that combines a Retrieval-Augmented Generation (RAG) pipeline with the TinyLlama-1.1B language model to deliver accurate, personalized, and real-time academic assistance to students. The system ingests domain-specific educational content, generates dense vector embeddings using Sentence-Transformers, and at query time retrieves the most relevant text passages to ground the language model's responses, effectively eliminating hallucinations that plague general-purpose AI systems.
Built on the FastAPI asynchronous web framework, EduMind AI provides a complete application stack including secure user authentication, session management, and an interactive chatbot interface. Unlike cloud-dependent commercial AI tutors, EduMind AI operates entirely on local infrastructure, ensuring complete data privacy and zero recurring API costs. Preliminary evaluations demonstrate 87.5% factual accuracy on undergraduate-level question-answering tasks, a 26-percentage-point improvement over retrieval-free baselines, with a user satisfaction score of 4.3 out of 5.

Introduction
The convergence of Natural Language Processing and web application development has created a transformative opportunity to reimagine how students receive academic support. Traditional educational systems rely on fixed instructor-to-student ratios, meaning the majority of students lack access to the immediate, personalized guidance that drives deep learning. While e-learning platforms have democratized content delivery, they remain fundamentally passive — a student watching a lecture video cannot ask a follow-up question.
EduMind AI addresses this gap by deploying an intelligent conversational agent directly within a web application. Students type natural language questions and receive coherent, factually grounded answers derived from curated educational materials. The system leverages a two-phase Retrieval-Augmented Generation architecture: an offline phase where documents are chunked and embedded into a high-dimensional vector space, and an online phase where query embeddings are matched against this index to surface relevant context for the language model.
The technology stack is built entirely on open-source components. FastAPI provides the high-performance asynchronous HTTP layer; Sentence-Transformers (all-MiniLM-L6-v2) provides the embedding backbone; cosine similarity provides the retrieval mechanism; and TinyLlama-1.1B-Chat provides the generative intelligence. This combination achieves commercial-grade tutoring quality with zero ongoing infrastructure cost, making it viable for institutions of all sizes.

Objective
The primary objectives of the EduMind AI project are:
• To design and implement a full-stack AI-powered educational tutoring web application using Python and FastAPI.
• To build a RAG pipeline that combines semantic vector retrieval with autoregressive language model generation for factually grounded answers.
• To integrate TinyLlama-1.1B as a locally deployable generative model, eliminating dependency on external cloud AI APIs.
• To develop a secure user authentication system with bcrypt password hashing, session cookie management, and protected routes.
• To provide a responsive, intuitive chatbot interface enabling seamless real-time student-tutor interaction.
• To evaluate the system on accuracy, retrieval relevance, response latency, and student satisfaction metrics.
• To demonstrate that compact open-source LLMs, when grounded with domain-specific retrieval, can rival commercial cloud AI tutors for educational question-answering tasks.

Block Diagram