chore: optimize AI models, improve PDF extraction latency, and clean dependencies

Implement 6-axis peer comparison and refined metrics

deploy: fresh start with clean history