Expert AI Trainer
Text
RLHFContributed to large-scale AI alignment and capability research initiatives, supporting post-training pipelines for frontier model development at a leading AI laboratory. Work focused on reasoning and alignment tasks for models evaluated on benchmarks such as MATH, and MMLU, specializing in mathematics and scientific reasoning domains. Core responsibilities included: - Authored long-form chain-of-thought (CoT) rationales for graduate-level mathematics problems to support model pretraining and fine-tuning pipelines - Performed RLHF annotation by evaluating and ranking model outputs, writing detailed preference justifications to inform reward model training - Extracted and structured reasoning traces from peer-reviewed mathematical and scientific literature to create high-quality supervised fine-tuning (SFT) data - Designed novel visual and spatial reasoning tasks with ground-truth solutions; conducted adversarial RLHF evaluations to benchmark model performance on geometric and diagrammatic inference - Executed QA/QC review across multiple concurrent data labeling projects, ensuring annotation consistency and adherence to task guidelines