Backend & streaming systems

I build for live video.

I’m Harry Dai. I build the backend systems that move, measure, and troubleshoot live video.

Santa Clara, California

LiveBridge RTMP

Browser simulation
No video selected

Choose a video to start.

Publisher

Ready
00:00

Your player

Broadcast ready
  1. Stream details
  2. Decoder setup
  3. First keyframe

Choose a video to start a broadcast.

Selected work

How LiveBridge works

Independent project

LiveBridge RTMP

I built a Go server that receives live audio and video from an RTMP publisher such as OBS and relays it to players such as VLC.

  • Go
  • TCP/IP
  • RTMP
  • FFmpeg
View source on GitHub
OBSLiveBridgeVLC
Parse the stream
I implemented the RTMP handshake, stateful chunk reassembly, and AMF0 commands for connecting, publishing, and playing.
Start playback midstream
When a viewer joins, the server replays cached metadata and codec headers. It holds audio and delta frames until the next video keyframe gives the viewer a decodable starting point.
Verify playback
I used FFmpeg to verify H.264 video and AAC audio playback.

Work history

Experience

TikTok

Jun 2026 – Present

Backend Software Engineer Intern

  • Consolidated separate livestream quality evaluation services into a real-time, RocketMQ-driven Go platform. Pluggable pipeline stages made it easier to integrate new evaluation methods and let business teams create workflows in a web UI without code changes.
  • Instrumented throughput, latency, and error rates with Prometheus. Built Grafana dashboards that combined pipeline and infrastructure metrics, and configured alerts for unexpected changes in evaluation volume, infrastructure issues, and excessive end-to-end latency.
  • Resolved a spike in end-to-end scoring latency by migrating overloaded pods. Used service logs and queue lag metrics to isolate the slowdown, then used CPU flame graphs to identify hotspots on the affected pods.
  • Built a ClickHouse-backed dashboard that lets business teams analyze average real-time livestream quality scores across all evaluation types by CDN vendor, region, and video quality tier.
  • Added audio to quality scoring by building an FFmpeg-based serverless function that detects abnormal silence in audio segments fetched from S3 and integrating it into the evaluation pipeline.
  • Go
  • RocketMQ
  • Prometheus
  • Grafana

TikTok

Jul 2023 – Aug 2024

iOS Software Engineer

  • Improved video-call logging and event tracking to diagnose a longstanding multi-device disconnection issue. Fixed it by matching signals from the long-lived connection to device IDs, reducing related complaints from five per month to zero.
  • Led an RTC optimization that adjusted video resolution to call window size and reduced downstream bandwidth costs by about $50,000 per month. Coordinated the rollout with the Android and server teams.
  • Refactored the input bar, action panel, chat backgrounds, and quick-access photo album from MVC into reusable components. Shipped them in an instant messaging SaaS SDK adopted by three other apps.
  • iOS
  • RTC
  • Messaging SDK

TikTok

Feb 2023 – May 2023

iOS Software Engineer Intern

  • Built an image loader for the quick-access photo album that shows low-resolution previews first and loads high-resolution images only for full-screen viewing or editing, eliminating white-screen flicker.
  • Shipped chat improvements that increased daily active users in chat by 600,000, including long-press background customization, input-bar button animations, and drag-and-drop media sharing.
  • iOS
  • Image loading
  • Chat UI

Get in touch.

I’m pursuing an M.S. in Artificial Intelligence at Northeastern University, with graduation expected in April 2027.