head of growth here

to tell you more about the process - getting good outputs required a lot of prompt engineering and numerous iterations using various models (we tested over 10 for this project).

the killer feature is structured generation (JSON mode) and streaming, which allowed us to get exactly what we needed in the right format and show users live generations of their personalities.

most importantly, it was easy to iterate on the prompts without messy abstractions or complex frameworks. our non-technical team members mastered the prompts, while an engineer focused on the code and making the app user-friendly.