Self-Improving Ad Copy Agent
Generates DTC ad copy, scores its own output with an LLM judge, remembers the winners, and gets better with every run.
Runs-·Golden-·Flagged-
How it works
01Generate
The agent turns a brand brief into three ad-copy variants: a headline hook, body copy, and a CTA.
02Judge
A separate LLM-as-judge scores each variant 1-5 on four dimensions: hook strength, brand alignment, clarity, and conversion intent.
03Remember
Outputs scoring >= 4.0 are promoted into a golden dataset and a vector memory; weak ones (< 2.5) get flagged for review.
04Improve & guard
Future runs retrieve those winners as few-shot examples, so quality compounds, and a regression gate blocks any prompt change that drops golden scores.