yeksax
← Back to home

Live Status

~15,000 active users

From minutes late to ahead of the broadcast.

When I took over Live Status, match updates arrived two to three minutes late, with inconsistent data and messages. I rebuilt the system so around 15,000 FURIA fans could follow matches on WhatsApp, with text, images, and video arriving up to 40 seconds before the broadcast.

Stack

language

TypeScript 01 language TypeScript Event shapes stay typed from socket to message. typescriptlang.org

back-end

Bun 02 runtime Bun Runs the finder and the listener, no build step in between. bun.sh Elysia 03 framework Elysia Serves the scoreboard the system captures for the match graphic. elysiajs.com Zod 04 validation Zod Parses the scorebot payload, which changes without warning. zod.dev

data

Redis 05 cache Redis Round buffer and the fingerprints that keep a play from going out twice. redis.io Firestore 06 database Firestore Match state and sent messages, so a restart resumes instead of replaying. firebase.google.com

infra

Google Cloud 07 infrastructure Google Cloud Where the listeners run, one container per live match. cloud.google.com Docker 08 packaging Docker One image per live match, so every listener starts identical. docker.com

integrations

n8n 09 automation n8n The bridge between the listener and delivery, editable without a deploy. n8n.io WhatsApp 10 channel WhatsApp The channel itself: text, image and video to every subscriber. business.whatsapp.com
01 capture

Match feed

The listener holds a live connection to the match's scorebot socket. Nothing depends on polling.

  • Finder schedules the match
  • WebSocket, no polling
02 context

Round buffer

Events accumulate until the round ends, in order and without duplicates.

  • Fingerprint deduplication
  • Score, sides and economy
03 delivery ~15,000 users

WhatsApp

AI writes the play and builds the media. Then it ships.

  • One push per round
  • Text, image and video
Read the full write-up →