LearnSphere AI

LearnSphere AI is an AI-powered “second brain”. A Chrome extension captures learning activity on YouTube, ChatGPT and Gemini, an LLM extracts the technical concepts, and the result is shown as an interactive knowledge graph (mind map). The goal is to keep everything you learn in one place without manual note-taking and to surface topics you are about to forget. Built by a four-person team over three Scrum sprints.

Key features

• Capturing learning content through a Manifest V3 Chrome extension
• LLM-based extraction of concepts, categories and difficulty; concept relations stored in Neo4j
• Interactive force-directed mind map
• RAG chat over personal knowledge using the Qdrant vector database
• FSRS-based forgetting curve, auto-generated quizzes, learning paths and topic clusters

Technologies

Python, FastAPI, LangChain, Google Gemini, sentence-transformers, Neo4j, Qdrant, React, Vite, Chrome Extension API, Docker Compose.

My contribution

• Sprint 1: scripts that generate synthetic learning data, and a Half-Life Regression (HLR) forgetting-curve baseline model (Python, pandas, NumPy, Matplotlib).
• Sprint 2: the FSRS memory engine: computing each concept's recall probability, visualising it on the map and updating it from quiz results.
• Sprint 3: topic clustering (the LLM assigns concepts to existing topics and the map groups nodes by cluster) and learning paths: Neo4j shortestPath routes for concepts already on the map, LLM-generated prerequisites for new goals.