Learnable: Adaptive Study via Error Reinforcement

Learnable is the study tool that actually remembers how you think — and where you break. Most AI tutors give you the same explanation every time. Learnable is different. Every mistake you make during a chat session gets embedded and stored in FalkorDB — a graph database purpose-built for connected knowledge. Over time, the system builds a map of your personal error patterns. Similar mistakes get linked automatically via vector similarity (SIMILAR_TO edges). When a deadline approaches, the AI doesn't give you a generic review — it reads your actual failure graph and generates a preparation plan targeting exactly what breaks down for you, not for the average student. The foundation is a directed knowledge graph built from your syllabus. Upload a PDF, a lecture recording, or paste a course description — Qwen3.5 397B A17B (running on AMD ROCm via vLLM on the AMD Developer Cloud MI300X) parses it and constructs a graph of lecture nodes, topic connections, and deadline nodes placed automatically on an interactive canvas. From there you navigate node by node. Each node is a full study environment: AI tutor chat in Theory or Task mode, generated flashcards, structured notes, and a Kanban board of practice tasks. The Planning Panel is a LangGraph-powered agent that lets you restructure the entire graph through conversation. Add nodes, connect concepts, generate tasks — all in plain language, with every change reflected instantly in MongoDB and FalkorDB simultaneously. Qwen's 72B vision model handles future canvas analysis. BGE-M3 powers all semantic search and error similarity. The result is a study platform that gets more useful the longer you use it — because it accumulates context about you specifically, not students in general.