From Diagnosis to Recommendation: Building an AI Ad Consultant in a Single Conversation
Session Overview
Building an AI ad consultant in a single conversation — can hours of ad performance diagnosis and consulting be compressed into one conversation?
This talk opens up the playbook behind a five-stage LLM pipeline that automatically diagnoses the root causes behind an advertiser's declining campaign performance and recommends the right resolution engine to fix it.
From data collection to a 25-branch diagnostic playbook, recommendation matching, report generation, and self-verification — see the full end-to-end pipeline, the hybrid rendering approach that suppresses hallucination (LLM production's toughest challenge), the eval verification loop that guarantees 100% numerical consistency, and real business impact and lessons learned from putting it into production.
Speaker
Yoonjeong Choi is Director of Ads Engineering at Coupang Ads, a veteran tech leader who has led large-scale engineering and ads tech organizations at global tech companies for over 20 years.
At Samsung Electronics, she led the launch of Samsung Ads' mobile DSP platform and built its machine learning-based data analytics infrastructure. She now leads engineering at Coupang Ads, directing a global organization spanning Korea, the US, and China across advertiser platforms, ad creative and integrity, content management, and large-scale ad tech architecture.
She heads Coupang Ads' core innovation initiative, "AI Ads Performance 10X," driving AI-powered advertising performance gains and spearheading the organization's transition into a Native AI organization. Guided by Coupang Ads' "Easy, Simple, Automation" philosophy, she builds AI-powered advertising automation solutions that drive sustainable business growth for advertisers of every size, including small business owners.