Research Assistant – LLM Simulation Evaluation and Optimization
As a Research Assistant, I designed and executed over 300 multi-agent LLM simulations to analyze language model (LLM) behavior under deceptive and adversarial prompting conditions. My work involved closely analyzing structured log outputs to identify hallucination patterns and rule violations. I tuned model parameters and optimized prompt engineering strategies to improve reliability and model performance. • Evaluated LLM outputs for hallucinations and rule adherence • Designed controlled experiments to assess prompt effectiveness • Iteratively adjusted temperature and sampling parameters • Documented findings and contributed to improved model reliability.