AI/ML Engineer for AI-News Video Pipeline automation

An end-to-end system that turns the day's AI news into finished YouTube videos (roundups + deep-dives) with almost no human touch — ingesting sources, clustering stories, writing a retention-tuned script, voicing it, and rendering the visuals. It runs a cost-optimized mix of LLMs (gpt-4o-mini for structural planning, DeepSeek for creative writing) with self-healing fallbacks, a Gemini production voice, and a motion-graphics-as-code render engine, all driven by a comedic, meme-forward "AI host by Eswar" editorial voice. The newest workstream adds emotion-first B-roll: a beat-level shot planner + a provider-agnostic asset-discovery module (Pexels/Pixabay + vision-verification + FLUX/local generation) that builds a reusable, metadata-rich visual asset library.