I built (mix of vibecoding and vibe-rage-debugging) an app for simulating focus groups using LLMs. A focus group is a small group of ppl hired to evaluate some product/concept/etc. The premise is that LLMs are, with the right prompting, good enough at imitating the probability distribution of opinions of common demographics if prompted to imitate them. Of course this depends a lot on the prompting -- if you prompt superficially the next-token predictor will return an equally generic answer.

You can feed in a website, define a study group and questions, and receive feedback. Currently in early beta.

Behind the scenes it uses a series of prompts to first figure out what your submitted content is about to prefill a suggested focus group and questions. If ticked, it will try creating "variations" of your content using an image model to be compared. Then the original and variations are shown to loads of instances of the openAI LLM with various prompts such as "You're a <demographic> study group participant evaluating <thing> so answer question <q>". Let me know if you have any suggestions for features etc and whether you find the concept interesting or would do something differently tech-wise. Thx!