Back to Systems
[SYSTEM_DEEP_DIVE]

Kinari Live

A real-time streaming system that enables creators to go live directly from the browser, delivering low-latency video to global audiences without relying on external tools.

Real-Time SystemsWebRTC / StreamingInfrastructure Design
[THE_PROBLEM]

Real-time video delivery is fundamentally complex. Traditional streaming pipelines introduce latency, break under network variability, and rely heavily on external encoding tools. The challenge was to design a system that enables real-time broadcasting directly from the browser, while maintaining performance, reliability, and scalability.

[01_IMPACT]

Zero-Tool Broadcasting

Real-time streaming directly from the browser, eliminating the need for OBS or complex external software.

Sub-Second Ingest

Near-instant transmission from creator to infrastructure.

Global Scalability

Seamless playback for thousands of concurrent viewers via automated CDN distribution.

Fault Resilience

Graceful degradation and adaptive bitrate handling to maintain uptime during network variability.

[THE_APPROACH]

Browser-Native Encoding

Using WebRTC to capture and encode streams directly from the client, bypassing external heavy-lifters.

Real-Time Transport

LiveKit SFU handles real-time distribution, ensuring streams remain fast and stable even as viewership grows.

Format Conversion

Egress pipelines to convert WebRTC to RTMPS for legacy CDN compatibility.

Global Delivery

Cloudflare HLS transformation for scalable, regional playback.

[SYSTEM_FLOW]
Creator
WebRTC
LiveKit
Egress
Cloudflare
Viewer
Ingest Latency~sub-second
Playback Latency~20–30s (Global HLS)
CapacityConcurrent viewers via CDN scaling
ResilienceAdaptive Bitrate + Fallback Handling
[02_CHALLENGES]

Latency vs. Compatibility

WebRTC offers ultra-low latency but struggles with massive concurrent global fan-out compared to robust CDNs. Extracting the pure WebRTC feed and dynamically transcoding it into HLS via egress workers was necessary to support unbounded audiences while keeping the host's ingest latency sub-second.

Network Variability

Browser-native capture is vulnerable to the client's network state. We implemented aggressive adaptive bitrate streaming (ABR) and simulated edge fallback mechanics to ensure that upstream packet loss resulted in resolution degradation rather than connection drops.

[TECHNICAL_STACK]
  • WebRTC
  • LiveKit
  • Cloudflare Stream
  • Next.js
  • Node APIs
[THE_BOTTOM_LINE]

Most streaming systems prioritize either low latency or scalability—rarely both. This architecture bridges that gap: creators get real-time responsiveness, viewers get reliable playback, and the system remains stable under growth.

"Building real-time systems requires more than connecting services — it requires understanding how data moves, transforms, and fails under pressure."

Building something that needs to scale reliably from day one?

Start a Project