Lab 14: Automated DPO Flywheel

Series: AI Eval Labs · Phase: Production Pipelines · Depends on: Lab 03

Core scope

• [ ] Capture simulated thumbs-down feedback
• [ ] Format feedback into preference pairs
• [ ] LoRA fine-tune a small open model and re-eval

Stretch

• [ ] Scheduled end-to-end loop with regression gate

Links

• Code: series/ai-eval/14-dpo-flywheel
• Demo: coming soon
• Write-up: coming soon on markrox.ai