Design a Telemedicine Video Chat

Hard45 min
1 / 31
System design answer

Design a Telemedicine Video Chat

Hard45 min31 sections
understanding•9 min read

Problem Statement: Regulated Real-Time Clinical Video at Internet Scale

Frames telemedicine as a compliance-bound, real-time clinical system rather than a generic video chat app.

Problem statement

Design a telemedicine video consultation platform that lets patients book appointments with licensed clinicians, join secure real-time video visits, exchange clinical documents (lab results, X-ray images), and optionally receive e-prescriptions routed to pharmacies. Every artifact that touches a visit — schedules, chat, files, recordings, logs — is ePHI (electronic protected health information) under HIPAA in the US, so encryption in transit and at rest, access control, and tamper-evident audit logs are architecture requirements, not features.

Why this is harder than a video chat app

A consumer video app can retry a dropped frame or lose a message. A clinical platform has three extra constraint classes. First, regulatory: HIPAA's Security Rule requires administrative, physical, and technical safeguards, a Business Associate Agreement (BAA) with every vendor that touches ePHI, and breach notification within 60 days. Second, clinical correctness: a prescription routed to the wrong pharmacy, an appointment booked into the wrong timezone, or a lab result attached to the wrong patient encounter are patient-safety incidents, not UX bugs. Third, real-time media economics: video dominates bandwidth cost and failure modes, and NAT traversal means a meaningful fraction of sessions cannot connect peer-to-peer at all.

Telehealth is operationally real and large. Teladoc Health reported 28.3 million visits in 2023 and a lifetime total above 350 million visits in its earnings materials; Amwell's Connected Sources powered 19.7 million visits in 2023; Zoom disclosed a peak of roughly 300 million daily meeting participants in April 2020, much of it driven by telehealth and remote work. These are public, self-reported figures and set the scale envelope, not the target of our fictional design.

The four planes of the architecture

  1. Clinical workflow plane: patient/provider identity, scheduling, intake, consent, visit lifecycle, prescriptions, billing hooks.
  2. Real-time media plane: signaling, SDP offer/answer, ICE/STUN/TURN, SFU forwarding, adaptive bitrate, quality telemetry.
  3. Data plane: encounter records, documents, ePHI storage with envelope encryption, retention and right-of-access workflows.
  4. Compliance plane: audit logging (tamper-evident), access policy, BAA inventory, breach detection, retention schedules.

A strong answer keeps the media plane fast and the compliance plane complete, and never lets one block the other: a TURN outage degrades video quality; a missing audit trail is a legal incident.

Design assumptions

Unless tied to a public figure, every number in this answer is an explicit design assumption: a mature network of 25,000 providers serving 6 million registered patients, 90,000 scheduled visits per day, 8% no-show rate, peak 22,000 concurrent video sessions across clinics, and 1.2 million document uploads per month. Scale numbers, SLOs, and retention windows below are stated targets, not vendor claims.

Key Highlights

  • •Every object tied to a visit is ePHI: schedules, chat, files, logs, and recordings all inherit HIPAA safeguards.
  • •Three constraint classes separate this from consumer video: regulatory (HIPAA/BAA), clinical correctness, and real-time media economics.
  • •Public scale anchors: Teladoc 28.3M visits in 2023; Amwell 19.7M Connected Sources visits in 2023; Zoom ~300M daily participants at April 2020 peak.
  • •Four planes: clinical workflow, real-time media, ePHI data, compliance — media speed and audit completeness must not block each other.
  • •Assumed scale: 25K providers, 6M patients, 90K visits/day, 22K peak concurrent sessions.
Say ePHI in the First Minute
Declaring that schedules, chat, files, and logs are all ePHI immediately reframes storage, logging, and vendor selection as compliance decisions. It separates senior answers from generic video-chat designs.
Do Not Design Zoom Without Context
A design that jumps straight to SFU topology without mentioning HIPAA, consent, or clinical correctness misses the point of the question. Media architecture is roughly one third of this answer.

Section Rescue Kit

Buzzwords to use:

ePHIBusiness Associate Agreement

Safe statements:

  • "Let me separate the clinical workflow plane from the media plane first, because their consistency and latency needs are opposite."
  • "Before choosing any vendor or codec, I will state which artifacts are ePHI and what that forces."
Design a Telemedicine Video Chat - System Design | WinJob | WinJob