I found an issue with the model used for data simulation. I used gemini 3.5 flash-lite for its speed for simulating posts from humans. However, the model routinely hallucinates and produces posts that are not very meaningful leading to confusing posts at times. I tested locally using gemini 4 argon and the models produce a much better result. Use the sample users to test out the app so that fewer simulated posts are generated.
Lesson learnt: Never skimp on model quality even if it's for simulating simple data
Log in or sign up for Devpost to join the conversation.