AI Copywriter

An internal AI content pipeline that lets anyone — no design or editing skills — ship on-spec content in one click.

Role Business owner + full-stack dev (vibe coding)

25min→2min per-cover design time
0 cover rework rate
-60~70% on sourcing & timeline assembly
≈150-180h person-hours freed / month

Problem

Intern design quality was uneven, and babysitting every revision was painfully slow. SOP docs alone couldn't enforce the standard — so I closed the technical gap with AI, hard-coding the spec into software and replacing daily manual review with tooling.

Approach

Self-adapting AI covers: embedded fonts with locked color / stroke / layout / safe-area parameters. The business side configures presets in an admin panel; operators apply and export in one click — cutting cover design from ~25 min to under 2, with zero rework. This module was ported back into the main commercial product.
Machine rough-cut + one-click CapCut import: an LLM splits scripts by narrative beat, sources footage across 1800+ sites and reviews frames; the program then deterministically assembles an editable CapCut draft.
A custom Electron “distribution helper” handles download → verify → path-rewrite → drop into the CapCut project as a frictionless one-click import, compressing the most repetitive step (sourcing / downloading / laying tracks) by ~60–70%.
Operable rules: creative specs (BGM / watermark / reference frame / compliance terms) apply line-wide via an admin “publish” button — no code changes needed.

Impact

At ~120 videos/month, the toolchain frees roughly 150–180 person-hours monthly — about one full-time operator's worth of capacity.
Scope: ~7 services and 40+ API routes, designed and built end-to-end solo.

Tech stack

Next.jsCloudflare WorkersD1R2Aliyun OSSGeminiPython (pyJianYingDraft)yt-dlpElectron