-
Group match movie result example
-
Create voting link and share with friends
-
24/7 Group concierge agent available to take modifications
What if personalization understood the group, not just you?
Most recommendation systems are built for one person: your feed, your music, your movies.
But a lot of life happens with other people.
At Mine is a social app for helping friends spend more time together in real life. For Showerhacks, we built Taste Match: a live group experience that finds the one film that works for everyone in the room.
How it works
The host starts a movie night and puts At Mine on the big screen.
Friends scan a QR code — no app required — and each add three films they love or import their Letterboxd favourites.
As people join, their names and favorite films appear live on screen.
Then At Mine builds a picture of the room's collective taste and searches for a film that works across the group.
Instead of simply optimizing for the average person's preference, we favor films that still work for the hardest person in the group to satisfy.
The goal: find the film nobody hates.
Then Usher, At Mine's AI host, reveals the match with a playful explanation of how everyone's wildly different tastes led there.
Finally, the host taps Plan a night with [film] and the recommendation becomes an actual movie night with invites, RSVPs, texts and push notifications.
How we built it
Firecrawl
Firecrawl gives At Mine its live film discovery layer.
When someone adds a favorite film, Firecrawl crawls outward through that film's TMDB recommendations.
That means the candidate pool isn't static. It grows based on the people actually in the room:
your favourites → related films → the room's possibility space
Firecrawl also powers our Letterboxd import and lets us discover films that aren't already in our index.
Moss
Films discovered by Firecrawl are indexed into Moss as semantic representations.
We create a taste representation for each person, combine those into the room's collective taste, and use Moss to retrieve candidates that are semantically compatible with the group.
We then rerank those candidates for group fairness rather than pure average similarity.
As new people join, the match can change with the room.
Usher
Our AI host, Usher, takes the strongest candidates and the group's tastes and turns the result into a personalized reveal.
Instead of:
"You should watch Arrival."
you might get:
"Davies wanted spaceships, Priya wanted feelings, and Alex apparently wanted everyone psychologically damaged. Somehow, Arrival works."
The recommendation itself becomes part of the night.
Challenges
The interesting challenge wasn't recommending a good movie.
It was defining what good for a group means.
A movie that one person loves and another hates isn't necessarily a great group recommendation, even if its average similarity is high.
So we designed Taste Match around group compatibility: reward candidates that perform well across everyone, particularly the person least aligned with the recommendation.
We also had to make the whole pipeline feel live: crawling recommendations, indexing films, computing the match and updating the big screen while people are standing in the room.
Caching and incremental indexing keep repeat interactions fast.
What we learned
Personalization gets more interesting when it becomes multiplayer.
Once AI can understand a group rather than only an individual, the same primitive can extend beyond movies:
- What should we watch?
- What should we play?
- Where should we eat?
- What should we do tonight?
- What would everyone enjoy on this trip?
Taste Match is our first experiment with that idea.
What's next
At Mine isn't trying to create another place where friends spend time online.
We want technology to make it easier for people to spend time together offline.
Most AI products help us spend more time with computers. At Mine uses AI to help us spend more time with each other.
Built With
- claude
- firecrawl
- flutter
- moss
- openai
- typescript
Log in or sign up for Devpost to join the conversation.