Reinforcement Learning from Human Feedback(RLHF) and model Aligment
Evaluated and ranked AI-generated responses based on factuality, coherence, and safety. Authored complex prompts to test model reasoning and creative writing capabilities. Performed SFT (Supervised Fine-Tuning) by rewriting model outputs to meet "Gold Standard" quality, directly contributing to the reduction of model hallucinations and improved instruction-following.