Video meetings

Participant count changes the architecture completely, and it is the first question worth answering. Peer-to-peer is genuinely simple and works well up to three or four people, after which every participant is uploading a stream to every other and the maths defeats a normal connection. Beyond that you need a media server relaying streams, which is real infrastructure with real cost — and the cost scales with minutes, so it is a business model question as much as a technical one.

Network conditions are the actual product experience. Someone will always be on a bad connection, and what the system does about it — dropping video before audio, reducing resolution rather than freezing, telling the user plainly that their connection is the problem — determines whether the call is usable. Audio is the part that must never degrade: people tolerate a frozen picture indefinitely and abandon a call within seconds of broken sound.

Recording and transcription are usually the commercial reason the product exists, and they carry obligations rather than just features. Consent has to be captured and visible, retention has to have a policy, and in many jurisdictions the rules differ by where each participant is sitting rather than where your servers are. Building the consent and retention model early is much cheaper than discovering it during a customer's security review.