How a short video app held a million users
A short video feed and live rooms in one app. It went past a million users after release, and the swipe never once asked anybody to wait for the next video.
- 1M+
- users on the platform after release
- 15
- of our team working on site in Dubai
- Company
- Undisclosed
- Industry
- Multi-sector group
- Company size
- 50 to 150
- Location
- Dubai, United Arab Emirates
Challenge
The brief was a short video app. What the client wanted was several products behind one login. The benchmark for every one of them had already been set by companies with thousands of engineers.
Several products inside one app
A short video feed and rooms for audio and video had to live in one product, with live chat and live streaming beside them. The app everybody compares this to carries none of it. There was no single product to copy, so every part of it had to be reasoned out instead.
The swipe has to feel like nothing
Users arrive with a benchmark already in their thumb. Any pause at all between two videos reads as a broken app rather than a slow one. The next video has to be ready before the finger has finished moving across the screen.
Rooms that hold a crowd
The audio rooms seat eight by eight. Video rooms run six by six and four by four, and every seat is sending and receiving at once. A room that degrades when the eighth person joins is a room nobody uses twice.
Live video on networks we do not control
A live stream gets judged on a phone and on a mobile network, never on office wifi. Bitrate and latency are the whole experience, and neither behaves the way it does on a desk. Getting it wrong is visible to every viewer in the room at the same moment.
Scale nobody could rehearse
The client was launching into a market they knew and could not say how fast it would arrive. A platform built only for launch week is a platform that dies in its second month. Build for a number nobody reaches and the budget that belonged in the product is gone.
A small client team, a large platform
The client runs businesses in several sectors and the internal IT team is a small one. Nobody there was going to absorb a platform of this size from documents and a weekly call.
What we built
Most of the work here went below the interface, where a user never looks and always notices. The answers run in the order the problems were listed.
One product, several pipelines
The feed needs one kind of plumbing underneath it, and a live room needs something else entirely. They share one identity and one social graph, so a user moves between them without meeting a seam. Only the parts that genuinely differ were built twice.
The next video is already there
The next video is fetched and decoded while somebody is still watching the one before it. The files are compressed and encrypted on the way through, and tuned for what a phone decodes cheaply. A large part of the engineering effort went into this one behaviour, because that is what it actually costs.
Rooms sized to the format
Audio rooms seat eight by eight, and video rooms run at six by six or four by four. The seat count is a product decision with a bandwidth bill attached, so it was set deliberately rather than inherited.
Streams that adapt to the network
Bitrate follows the connection the viewer actually has rather than the one we hoped for. Latency was treated as a budget to be spent, and every part of the path had to justify its share. The result holds up on a phone, because a phone is where every part of it was tuned.
Capacity that follows demand
The platform grows with the load rather than sitting at a size somebody guessed in advance. That is the only version of this that survives both a quiet launch and a loud one. The million users that arrived turned out to be an operations day rather than a rebuild.
Fifteen of us in the room
Day to day the work ran remotely, which is how most of it gets done. For the parts that needed a whiteboard we sent fifteen people to Dubai, project managers included, and worked in the client's own building. Some things get settled in an afternoon there that would have taken a fortnight of email.
What changed
The app launched, the audience arrived, and the platform did not have to be rebuilt to hold them.
Past a million users
The platform went past a million users after release and behaved the same way it had on day one. Scaling turned out to be a question of capacity rather than a second version of the product.
A swipe nobody thinks about
The feed moves the way people expect it to move. Nobody using it has any idea how much work sits under that, which is the correct outcome for this kind of feature.
One app doing the work of several
To the person holding the phone, the feed and the live rooms are the same app. The client sells a single app rather than a suite nobody wants to install twice.
Let AI do the repetitive
half of the job.
Data entry, answering the same tickets, chasing numbers between systems. We automate the parts that repeat. Your team keeps the parts that need judgement.
