Radio for the factory floor
A radio in the phone for a loud factory floor. Hold the headset button and talk, and everything that's said turns into Swedish text.
- Client
- A manufacturer in Småland
- Since
- 2026
- Status
- In development
- We built
- iPhone app, web app, on-prem speech recognition
Overview
A factory floor is loud, and whoever needs help at a machine can't walk around looking for the right person. The manufacturer wanted anyone to be able to reach everyone, all the time, by voice. So we built a radio that lives in the phone, with a headset in your ear.
The challenge
The factory floor is loud, and the manufacturer wanted anyone to be able to reach everyone, all the time, by voice.
What we built
A radio in the phone: the headset button, channels per area, all-call for supervisors, and Swedish text of everything that's said, with speech recognition on their own servers.
You log in once and you're straight into the channel for the whole factory and the one for your own area, such as assembly, warehouse, or maintenance. No menus. To talk, you hold the button on the headset and let go when you're done. One person at a time per channel, like a real two-way radio, and 30 seconds at most per call.
Everything that's said turns into Swedish text as it happens. The speech recognition runs on the manufacturer's own servers, not in somebody's cloud. Every call is kept as audio and text and can be played back, so whoever missed something over the noise can read it afterwards. They decide how long it's kept, and the app tells everyone that calls are recorded.
Phones without a headset stay quiet by default, so nobody's pocket starts shouting across the hall. Supervisors have an all-call that overrides every channel and turns the speaker on anyway. Anyone at a desk, or at home over VPN, runs the same app in the browser and holds Space to talk.
How we built it
Everything starts from the person at a machine with a headset in their ear: one button to talk and no menus. The screen is there for whoever wants to read what was said.
Channels that follow the factory
One for the whole floor and one per area, and you land in the right ones the moment you log in.
The button on the headset
Hold to talk by default, or tap once to latch as a setting. Start and stop tones, a busy tone when someone else has the channel.
Swedish text in real time
Speech recognition on their own servers, with every call kept as audio and text for as long as they choose.
All-call for supervisors
It overrides every channel and turns the speaker on, even on phones without a headset.
Presence
Who's in, in which channel, on a headset, in the browser, or on a phone, and who's talking right now.
One login
The same account as their other systems, and the role decides who can make an all-call.
One interface for phone and desktop
The iPhone app and the browser show the same app.
- SvelteKit
- .NET
- TypeScript
- Python
- PostgreSQL
- WebSockets
- Tailwind CSS
- Docker
- Kubernetes
Gallery
Key metrics
- Button on the headset to talk
- 1
- Hold it to talk and let go when you're done. Or tap once to latch, if that's how you've set it up.
- Longest call before it cuts
- 30 s
- A tone warns you at 20 seconds. One person talks at a time per channel, so nobody can block it for long.
- Cloud services that hear what's said
- 0
- Speech recognition runs on the manufacturer's own servers, and they decide how long audio and text are kept.
Got a system everyone's waiting on?
Tell us what you want to build and we'll take it from there.
Next project
Property
