AI-asistovaná tvorba znalostních map

Hlava, Ondřej

Search engines and LLM chatbots retrieve and summarise information well, but nei- ther helps a learner see how a topic is shaped or how its parts relate. MapTopics is a web application built to bring those two worlds closer together inside a shared spatial medium: an interactive, source-grounded mind map that the user can grow, reshape, and personalise with AI assistance. The canvas is deliberately open. The user drags, draws, and annotates as they would on any whiteboard, and calls on AI only when it helps - to draft an overview of a new topic, expand a node, rephrase one for a different audience, or answer a question about it. A personalised feed and a responsive search suggest where to go next; a local-first library keeps everything the user makes on their own device, with optional cross-device sync. The system is built around a pluggable generation pipeline, a hybrid recommender that combines lexical and semantic-embedding similarity with collaborative signals and a measure of serendipity, and a small evolutionary layout step that keeps generated graphs readable. The whole stack runs offline against a small local language model, which is enough for a wide class of AI-assisted tools and avoids the latency and privacy cost of hosted inference by default. A handful of experiments validate the core defaults...

guest :: login Digital Repository
		Search		Submit		Help		About