SmartFlix

Dense, complex documents are notoriously hard to learn from — walls of text, technical jargon, and abstract concepts create a steep barrier for learners. Manually creating high-quality educational video content is prohibitively expensive and time-consuming, often taking days per video. The core challenge was fully automating the pipeline from raw document ingestion to a polished, pedagogically sound video — without sacrificing the teaching quality that makes content actually stick.

We designed and built a 6-phase AI agent pipeline: Pre-processing extracts and normalizes content from any format. Understanding performs deep semantic analysis to identify core concepts, relationships, and complexity levels. Planning creates a pedagogical strategy with Feynman-style simplification. Generating produces the narrative script with visual element markup. Polishing runs quality loops to refine clarity and pacing. Voice Selection matches the optimal TTS voice to the content style. Remotion 4 handles programmatic video rendering, BullMQ manages the async job queue through Redis, and MinIO provides S3-compatible asset storage. The entire system runs as a Turborepo monorepo with separate web, API, and worker services.