How I Built an AI Study Buddy That Generates Notes, Tutorials, and Self-Validated Tests

This article documents a multimodal AI study pipeline built on NVIDIA Nemotron Omni and vLLM that converts textbooks, lecture videos, handwritten notes, and study-group chats into three synchronized outputs: organized notes, worked tutorials, and calibrated practice tests. The key technical idea is a self-evaluation filter where the same model both generates and validates questions, rejecting ambiguous, weakly grounded, or low-confidence outputs before they reach students.