Tutor: An Adaptive AI Learning Platform Based on RAG and Knowledge Graph

Tutor is an open-source adaptive AI learning platform built with C#/.NET Blazor. Its core goal is to automatically convert traditional books and documents into navigable interactive courses, leveraging LLM, RAG, and knowledge graph technologies to provide personalized learning experiences. Key features include multi-format document parsing, semantic search, dynamic course generation, and progress tracking.

Developed by the mindattic team, Tutor uses the C#/.NET tech stack. It aims to transform traditional learning materials into structured, interactive courses. By integrating LLM, RAG, and knowledge graph, it enables personalized learning paths for users.

Multi-format Document Parsing

Supports PDF, EPUB, HTML, DOCX, and legacy formats (like .doc, .rtf). Uses two-stage parsing: managed C# for modern formats, and LibreOffice/Calibre tools for legacy ones.

LLM Service Routing

Flexible LlmServiceRouter supports OpenAI, Claude, DeepSeek, Gemini. Uses MindAttic.Legion library for unified authentication, retries, and circuit breaking.

Concept Extraction & Knowledge Graph

ConceptMapService extracts JSON concepts from document fragments; KnowledgeGraphService links concepts across resources; OrphanConceptLinkerService reconnects isolated concepts to form a knowledge network.

Semantic Search & Vector Storage

EmbeddingService + VectorStoreService enable semantic search, allowing natural language queries to find relevant content (not just exact matches).

LSH/SimHash Deduplication

SimHashService generates content fingerprints; LSHService quickly identifies near-duplicates. This ensures unique concepts in the knowledge graph and avoids redundant information.

Dynamic Course Generation

CourseStructureService converts concept maps into hierarchical learning paths; SectionContentService fills chapters with source material, automating course creation from any document.

Progress Tracking & Smart Quizzes

Tracks user learning progress and provides LLM-driven quizzes. Quizzes adjust difficulty dynamically based on user responses for adaptive learning.

Modular Deployment

Modular architecture includes Core library, Shared components, Blazor Server app, and Cli tool—same logic works in web and command-line environments with unified course data.

Application Scenarios

• Educators: Fast digitization of traditional textbooks.
• Self-learners: Personalized learning paths.
• Content creators: Automated pipeline for knowledge products. Demonstrates deep RAG application in education, showing AI can understand and restructure knowledge.

Tutor represents cutting-edge AI education tech, forming a closed loop of learning content production and consumption via document parsing, semantic retrieval, knowledge graph, and adaptive learning. Its open-source nature provides a valuable reference for the edtech community.