The problem
Gamers are one of the largest communities on the internet, but the platforms they use don't know what you play, can't recommend what you would play next, and treat a fourteen-year-old the same as an adult. Most teens play games, and a lot of them run into harassment doing it.
GamerTag was one product for those three gaps: identity around what you play, discovery from that, and safety designed in, not moderated after. I co-founded the company with a CEO and a CDO and ran product and technology. We started in Malta, then moved to the UAE.
Role
I owned product and technology strategy, the design system and brand, the cloud architecture, and the technical direction of the AI platform. We had pre-seed funding across three rounds. I hired into a company of around twenty and led an engineering team of about ten across mobile, backend, and ML, while I stayed in the codebase on both sides. More than 70% of the product, brand, and campaign design was mine, delivered to the team in Figma and the rest of the toolchain.
Design
Gaming social is crowded, and most of it looks the same: dense, functional, and visually a decade behind. We had no distribution and no network to start from, so the first decision was where to spend early capacity. We spent it on craft.
The brand language came from something gamers already recognize: simulated lighting, glow, and reflection, not a generic dark UI, plus custom type. Perceived speed was part of that brand too. Blurhash placeholders and skeleton states were designed with the screens, because on mobile in Lebanon or Jordan, how fast the app feels is the product.
That work won the Red Dot Award: Communication Design 2024, where I'm credited as design lead.
Safety
Protecting younger players was part of the product, so it couldn't depend on a moderation team we couldn't afford to scale. Age had to live in the system, not in a policy doc.
Birthdate is captured at onboarding, carried as a claim in the auth token, and checked against a content tier that every post earns on ingest, before it can reach a feed. The pipeline reads text, reads images, and reads text inside images, so abuse can't hide in a picture. Anything that fails leaves timelines automatically and lands in a review queue.
Every post pays for classification before it is visible, which costs latency and infrastructure. The benefit was that safety held at 23,000 monthly users without a moderation department, and would have held at ten times that.
Platform
I set the technical direction for a serverless consumer platform on AWS: a GraphQL API, a Flutter client for the full product, and an event-driven AI layer that classified and ranked content as it moved through the system. Three environments sat behind automated delivery, so the team could ship without the platform becoming the bottleneck.
Our CDO, a computer science PhD candidate and AI researcher, led the recommendation algorithm design. I owned the serverless infrastructure those models ran on: the event contracts between services, the ingest path that fed them, and the database-stream pipeline that kept user and content vectors current under real traffic, not overnight batches.
Ads
The same signals that ranked the feed could target a paid post, so we designed and shipped Promotions, a self-serve ads product inside the app, live on the App Store and Google Play. Any account could boost its own post with AI-assisted audience targeting and an estimated reach before paying. It shipped in our last month: finished and built properly, but it never got the time to earn.
Grow
A consumer product also needs the commercial layer, not only the screens. We ran acquisition in-house on Apple Search Ads and Google Ads, with attribution and BI feeding spend, and we used the same in-house production for a sponsored Dakar 2024 rally livery and stand design at LEAP and NorthStar Dubai.
Under that, I built the operating layer most people never see: internal dashboards for spend, revenue, metrics, and moderation, mailing, and GDPR done properly with an outside firm, including the data modeling and lawful basis for AI use, not just a consent banner.
Sprint performance for a team of ten lived across Jira tickets, estimated points, and a handful of spreadsheets, with no single place to see how a sprint actually went. I built and deployed gamertag.team, a web app on its own short domain, with secure auth, wired into our Jira tickets to track task time against estimated points. That gave us one source of sprint data, so retrospectives could use real numbers instead of gut feel.
Outcomes
We launched into EU and MENA and reached up to 23,000 monthly active users, at a two-and-a-half minute average session, from a standing start, with no incumbent distribution and no existing audience. Strongest adoption came from Lebanon, Jordan, Azerbaijan, Italy, and Georgia, a mix we hadn't predicted, and we reoriented toward it once the data arrived. A seed round opened in 2025, and GamerTag became a registered EU trademark.
In hindsight
We shipped late because the product was too broad for the stage we were in. Feed, chat, nearby discovery, a huge game catalog, AI recommendation, safety, ads, GDPR, and infrastructure built for scale were all defensible on their own, but together they became the go-to-market by default. We proved the platform could carry far more users before we proved the habit loop was strong enough for the first ten thousand. A narrower launch, built to the same standard and released earlier, would have taught us more.
This was also built before agentic coding tools. The same platform today would take a fraction of the team, the capital, and the calendar, and that is the lesson that most changes how I would resource a product now.
The founding team came apart under funding pressure and visa and relocation complications, and the company wound down in 2026.