Context
At MTS Link I’m one of the two core developers of the client WebRTC library in the Media Core team — the media layer behind video calls in the browser.
Problem
When a participant lost their network, getting back into the call took about 30 seconds. In a live conversation that’s a long time — the person simply drops out of the discussion.
Why 30 seconds is a long time
Until the connection is back, the participant can’t see or hear anyone, and nobody can see or hear them. Reconnection time is time spent outside the conversation, even if the network came back after a second. This is a general explanation, not a description of the company’s code.
What I did
Reworked how the library recovers the connection.
Result
Reconnection now takes 6 seconds instead of 30 — five times faster.
Related work on the library
- Fixed the “black screen” bug and a connection leak — a top recurring Sentry error is gone.
- A second connection mode — MCU (Vinteo) alongside SFU.
- Quality presets served from the backend without a frontend release; batched telemetry through a single instance.
- Linting and git hooks, unified error handling, ~110 tests, integration docs.
Implementation details are the company’s internal code. Happy to talk through the approach in person.