Hi Elvoro team — I read your Python AI Creative Engineer brief carefully. Your exact operating philosophy (Claude Code / Codex / xAI + Python/FFmpeg pipelines + ruthless human visual taste to kill AI slop and track cost-per-approved-post) is the exact production system I already run daily for US eCommerce brands.
Built an automated Claude + Codex + browser QA factory that upgraded thousands of product creatives, enforced aspect ratios, caught brand/product inaccuracies, and visually verified desktop + 375px mobile renders.
Engineered multi-source scraping, structured catalog enrichment, retry watchdogs, and token/cost budget controls so pipelines never stall silently overnight.
Every creative batch passes programmatic checks (FFprobe 9:16 ratio, audio stream, hook length, cliché blacklist, cost-per-approved-post) followed by side-by-side visual review.
Before reaching out, I ran an automated visual + DOM check on elvoro.com:
alt attributes — an easy SEO & accessibility win.

creative_qa_gate.py (Python + FFprobe)To demonstrate code style rather than just talk about it, I wrote a working Python + FFprobe gate that scores opening hooks, flags AI-slop phrasing, verifies 9:16 vertical specs & audio streams, and calculates cost per approved post:
{
"pipeline": "Elvoro Anti-AI-Slop Creative QA Gate (Prototype by Zach)",
"total_assets_tested": 3,
"approved_count": 2,
"rejected_count": 1,
"total_generation_cost_usd": 1.54,
"cost_per_approved_post_usd": 0.77,
"rejected_example": "reel_v1_slop.mp4 -> AI-slop clichés + 18-word bloated hook"
}
Proposed Kickoff: Happy to start immediately with a scoped paid sprint (e.g. auditing & tightening one failing creative pipeline or building a new short-form hook/FFmpeg assembly workflow) directly with the founder, async-first.