PostMortemEng

PostMortemEng

A multi-agent AI video pipeline that turns forensic-engineering research into narrated YouTube Shorts, cutting end-to-end production from 6–10 hours down to about 7 minutes per video.

Architecture & Approach

A four-stage agent pipeline — Script Writer, Art Director, Component Builder, then a QA Reviewer — where the QA stage acts as a schema-validated gate between every handoff, rejecting bad generations before they become broken videos (20 rejected across 7 test renders). Narration runs through Kokoro TTS with Whisper for word-level caption alignment, and Remotion + GSAP handle rendering and motion. Render time was the real bottleneck undermining the "fully automated" claim; tracing it to a worker-concurrency misconfiguration and fixing it delivered a 24× render speedup, which is what made the 7-minute number real.