Case study №02
Hidnn: dev, PM & QA
Owning a privacy-first dating app end to end
- Role
- Dev · PM · QA
- Timeline
- 2025 — Present
- Scale
- Flagship dating app
Context
Hidnn is a privacy-first dating app — built for people who want to meet someone without broadcasting that they're looking. On a platform of thirteen apps run by a small team, nobody gets to be just an engineer. On Hidnn I ran the whole loop: writing the features, deciding what ships, and breaking it before users do.
The dev hat
A privacy-first dating app lives or dies on trust, so the deepest engineering work went there. Banned users don't leave — they come back with a new number. Hidnn is wired into a device-fingerprinting fraud pipeline that recognizes the device itself, so a ban survives re-registration. And because every profile photo is a liability until proven otherwise, uploads run through an AI moderation pipeline — a multi-provider vision-LLM architecture, benchmarked across 6 models to get cost down to ~$0.30 per 1,000 images. The rest of the hat: monetization surfaces wired to payment cohorts, and feature flags so risky changes ship dark and roll out in layers.
The PM hat
Deciding what ships and when: reading funnels in ClickHouse, running A/B tests on onboarding and paywalls, and turning behavioral data into a roadmap. The discipline this teaches is altitude control — the same person who knows how expensive a feature is to build also knows exactly what it's worth to users, so “should we?” conversations get honest fast.
The QA hat
Every release walked through device passes on real Android and iOS hardware before it reached the stores: regression checklists for auth, payments and notifications, plus deliberately hostile testing of new flows. Owning QA for your own code changes how you write it — you stop shipping things you wouldn't want to test.
What it produced
- 350+
- banned users blocked from returning
- ~$0.30
- to moderate 1,000 images
- 0
- handoffs between build, ship, test
In dating apps, trust is the product — a user who meets one fake profile doesn't come back. Fraud and moderation are usually the problems teams get to after growth; on Hidnn they were built into the platform itself. And the loop between “a metric moved” and “the code changed” is as short as it gets: the person watching the funnel wrote the screen.
Next case study
Sehat: built for my parents