Parastore - Simulate real store with LLM-powered synthetic consumer

Hey Product Hunt! 👋 I'm Kay from Intellicia. We've been building AI synthetic consumer technology for the past year — helping brands like CJ, Pulmuone, and Fursys test products and messaging with AI personas instead of traditional surveys. Parastore is a different beast. Instead of answering surveys, our synthetic consumers now physically walk through stores. They browse, pick up items, impulse-buy snacks near the checkout, and generate revenue data — all in a 3D simulation. We're open-sourcing the entire simulation framework (MIT license) because we believe agent-based behavioral simulation is a space that deserves more builders and researchers. A few honest notes: The validation numbers come from our proprietary persona engine. The OSS version is a simpler pipeline — still useful for layout testing and agent behavior research, but don't expect the same accuracy out of the box. Each sim run calls the LLM hundreds of times, so it's not free to run. This is a simulation tool, not a crystal ball. It shows plausible outcomes, not predictions. Would love your feedback — especially from anyone working on LLM agent simulation, retail optimization, or behavioral AI. ⭐ Star us on GitHub if this looks interesting!

@kaylim022 this is so cool! Out of curiosity - did you observe “unconventional” agent behavior that was far off from what humans would do?