PersonaGen

PersonaGen is a framework for generating and evaluating long-form, timeline-grounded synthetic personas. Rather than treating persona creation as a single prompting task, the system uses a staged pipeline that builds structured seeds, explicit timelines, outlines, long-form sections, repair passes, and audit artifacts. The slides emphasize that the main challenge is not producing one plausible persona, but producing many personas that remain distinct, coherent, temporally consistent, and measurable at scale. Current results suggest that frontier models can produce strong long-form persona artifacts, but quality must be interpreted alongside diversity, repetition, reliability, runtime, and failure modes.