What if you could see user reactions before launching?
hey everyone 👋
been building something called Lumina — https://www.getlumina.in/
The idea is simple: instead of launching and then figuring out what works, you can simulate how people react before going live
You can test ads, landing pages, copy, or even raw ideas against AI personas and see:
What grabs attention
where people get confused
objections they have
What might actually convert
still very early, figuring things out in public
Would love honest feedback if you try it 🙏
Also started a small community for builders/marketers testing ideas early:
👉 join here: https://discord.gg/aaq8Y86kfj
Comments (4)
This is a very interesting idea! Although if it is AI simulation, then AI will most likely be more successful in figuring out UI or what to do next, than when a real human user tries.
Hey Piyush! Testing against AI personas before launch is a cool concept. Especially for solo founders who don't have budget for real user testing early on. How close are the AI persona reactions matching what people actually see post-launch? Would love to know if you've been able to compare the two yet.
What we’re seeing so far is that Lumina is quite good at picking up: - clarity issues - confusion points - common objections - relative performance between variants Where it’s still evolving is matching absolute outcomes (like exact conversion rates), since real-world distribution, timing, and external factors also play a role. We’ve started doing some early comparisons between simulation results and real campaign feedback, and directionally it’s been promising, especially for identifying what won’t work. The goal isn’t to replace real users, but to give you a strong signal before you spend.
Simulating reactions before you launch is such a good instinct, most founders (myself included in FounderFlow's early days) find out what's confusing about a landing page from real users after it's already live, which is the expensive way to learn. The part I'd want to know before trusting the output: how do you keep the AI personas honest instead of just telling you what you expect to hear? That's the same trust problem I ran into building FounderFlow, an AI that's optimized to sound helpful can drift into telling you what you want rather than what's true, so I ended up building explicit confidence grading into every insight rather than trusting a single clean-sounding answer.
Sign in to comment or upvote.