Most "AI study tools" are the same thing wearing a different skin: dump in a PDF, get flashcards or a chatbot that just answers questions. PrismLearning.AI is built to actually teach. Upload a PDF, PowerPoint, or YouTube video and it's restructured into a chapter-by-chapter curriculum, taught by Lumi, an agentic tutor that returns structured JSON instead of raw text — so the AI physically drives the UI: scrolling to the right concept, unblurring locked sections as mastery is proven ("Fog of War"), and spawning flashcards mid-conversation. It won't just hand over answers; it scaffolds, asks the student to teach concepts back, and lets you talk to it directly through voice. Assessment goes deeper than multiple choice: quizzes include real math (LaTeX-rendered) and code questions, graded properly, with a confidence check before every reveal so students learn to notice their own blind spots. Flashcards run on real spaced repetition, and mastery decays over time if you don't come back — progress reflects actual retention, not a checkbox. Mastering a chapter for the first time triggers a themed, timed boss battle exam, with XP, levels, and streaks turning study into something worth returning to. Under the hood: tutoring runs on gpt-oss-120b via Fireworks AI's serverless platform, powered end-to-end by AMD Instinct GPUs. Flashcard generation runs on a second, smaller model — Gemma 3 27B — deliberately matched to a short templated task instead of over-provisioning the same large model everywhere. A gated Enterprise path (Settings → AI Model Provider) proves the same architecture can flip tutoring itself onto a dedicated Gemma 4 deployment for organizations that need full data residency. Built solo by a first-year BSIT student as a first real SaaS attempt — full demo video, live deployment, and documented architecture included.
Category tags: