Clark

Hey PH

We re building LoveActually.ai, an AI matchmaker launching soon.

I wanted to share a technical dilemma we ran into and hear how other builders think about it.

Hey Product Hunt,

I m the maker behind LoveActually.ai. During our beta, we made a controversial product decision that split our team, and I want to hear your take.

Most AI companions (like ChatGPT or Pi) are RLHF-tuned to be supportive, polite, and agreeable. But in the dating world, we found that "politeness" was actually hurting our users. They didn't need a cheerleader; they needed a wake-up call.

So, we built Astute Kitty. We engineered its system prompt to prioritize "rational critique" over "emotional safety." It calls users out when their dating standards don't match their own profiles.

The hardest part of building an AI matchmaker wasn t the model.
It was deciding what not to optimize.
From an engineering perspective, it s very easy to make a dating product look alive :
more swipes, more matches, more notifications.
The harder problem was building systems that could:
tolerate silence
slow down recommendations
and still be confident in a match
We spent a surprising amount of time designing memory and decay:
what the system should remember about a person,
what it should forget,
and when it should wait instead of acting.
Astute Kitty doesn t just score compatibility
it reasons about timing, emotional readiness, and communication rhythm.
Loving Kitty exists for a simple technical reason:
emotional context doesn t fit neatly into state machines.
Building this forced us to treat users less like traffic
and more like long-running processes with history.
It s slower.
It s less impressive in dashboards.
But it feels more honest.
LoveActually wasn t built to feel busy.
It was built to feel calm.