lookatstudy

LookatStudy is a desktop application that converts almost any learning material into a structured, gated course. It accepts GitHub repositories, local folders, web articles, arXiv papers, video links, pasted text, EPUB books, and audio recordings. Supported document formats include Markdown, Jupyter notebooks, PDF, PowerPoint, HTML, and plain text, while code files across many languages are treated as teaching material with docstrings as prose. Audio and video sources can be transcribed locally using Whisper, and bilingual content is paired automatically. The app organizes imported content into a skill map where sections and lessons become nodes on a path. Lessons unlock sequentially, and each lesson is broken into knowledge points. Mastery of a lesson is determined by the weakest knowledge point, so users must demonstrate understanding of all concepts before progressing. Chapter exams act as gatekeepers, with timed questions generated from the chapter's knowledge points. Exam results are broken down by knowledge point to guide review. An AI tutor tracks per-concept mastery using a Bayesian Knowledge Tracing (BKT) model. The tutor identifies weak areas and focuses questions and explanations on concepts the user struggles with. Users can switch teaching styles between direct explanation, guiding questions, or self-directed practice. The AI cannot modify learning records without user approval; it drafts proposal cards that require confirmation. Chat threads run asynchronously, allowing multiple conversations to stream simultaneously. Spaced repetition uses the SM-2 algorithm to schedule reviews before forgetting occurs. Daily XP, streaks with freeze options, and a mixed review drawer provide motivation. All data is stored in a single SQLite file locally, with no account or cloud sync. Users supply their own LLM API key, with nineteen preset providers and support for any OpenAI-compatible endpoint. The key is isolated in the main process and inaccessible to the renderer. The app includes text-to-speech with sentence highlighting, offline neural voices, and dictation via local Whisper. A small interactive robot companion reacts to user actions and can be customized. A built-in guide course demonstrates the full learning loop without requiring an API key. Installers are available for Windows, macOS (Apple Silicon), and Linux, with a mobile path via Termux on Android. A plugin version exists for DeepSeek Harness, providing the same study features within that agent framework.