Context
A matchmaking platform is not judged on how many features it has, but on two points: how relevant its suggestions are, and how much trust it inspires. A user shown profiles unrelated to their criteria leaves within minutes. A user exposed to unmoderated behaviour does not come back at all.
The problem
Both difficulties are technical before they are functional.
- Affinity scoring has to stay fast while crossing several criteria and a geographic distance
- Messaging has to feel instant without holding open connections for inactive users
- Moderation has to act quickly, without turning the team into a permanent watch service
- The data handled is personal, sometimes sensitive, and the regulatory framework applies in full
The solution
An application built around a rich but quick-to-fill profile, a suggestion engine and real-time messaging.
- Profiles with photos, interests and search criteria
- Swipe navigation, with suggestions prepared ahead so nothing is waited for
- Compatibility score crossing declared affinities with geographic proximity
- Instant messaging opened only after a mutual match
- One-gesture reporting, and moderation tools on the admin side
My role
Design of the data model, the suggestion algorithm, the API and the interface, plus the real-time layer. I also built the moderation tooling, often left until last although it determines whether the product is viable.
Constraints
- Personal and sometimes sensitive data: minimisation, retention periods, right to erasure
- Affinity scoring has to stay tractable as the number of profiles grows
- Real-time connections held only for genuinely active users
- Fake profile prevention, without turning sign-up into an obstacle course
- Photos served quickly, on mobile, without visible quality loss
Outcome
An application where the main journey, discover then talk, takes a few gestures, and where moderation has the means to act.
- Suggestions are prepared ahead, so navigation stays fluid
- Messaging opens only after mutual consent, which limits harassment at source
- Reports arrive with the context needed to make a decision
Outcomes are described by the capability delivered. No commercial performance metric is claimed here: usage figures belong to the client, and I do not publish numbers I cannot substantiate.