How it works: - Ingest any file (PDF, markdown, images, handwritten notes via vision LLM) - Extract concepts and generate embeddings - Project to 2D with UMAP, cluster with HDBSCAN - Render as interactive honeycomb grid - Auto-generate cloze + Q&A flashcards - Schedule reviews with FSRS algorithm

Stack: FastAPI, LangChain, sqlite-vec, Nuxt 4, D3. Works with OpenAI or Ollama (fully local).