AI Model Training

Shaping the next generation of AI models

Contributed to the training and evaluation of frontier large language models through structured human feedback for multiple frontier AI labs. The work involved multi-axis rating systems used for supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF).

Post-training evaluation: adversarial prompt design, rubric construction, golden-response authoring, and reasoning-trace evaluation across text, audio, and multi-turn systems — reviewer-rated top-tier for producing clear model failures.

Contributing to the next version of a flagship AI assistant

Evaluated and rated outputs for a new version of a major AI lab’s flagship conversational assistant, applying a rigorous multi-axis assessment framework. The work focused on improving the model’s conversational quality, factual reliability, and alignment with user intent.

Evaluation work across the teams behind leading frontier models

Worked on evaluation tasks for another frontier lab’s assistant and copilot products, rating model outputs across multiple quality axes. The project involved assessing the model’s ability to follow complex instructions, provide grounded and factually accurate responses, and maintain safety guardrails.

Beyond direct evaluation, also served in a quality assurance capacity — reviewing other evaluators’ ratings for consistency, accuracy, and adherence to rubric standards. This meta-review role helped maintain the integrity of the training data pipeline.