Case Study 01
ChatsX: one real-time inbox for Meta messaging channels.
A multi-platform chat aggregation system merging Facebook, Instagram, and WhatsApp conversations into a single operational inbox designed around 20K WebSocket connections, Meta events, and LLM-assisted agents.
The brief
Messaging systems become backend systems very quickly.
The work required real-time delivery, account-level state, Meta Graph API integration, queue behavior, and enough operational clarity for support teams to trust the inbox at high connection volume.
The LLM layer added another constraint: agents had to use conversation patterns, suggest replies, and improve support flow without making the core messaging workflow fragile.
20K
WebSocket connections designed into the real-time inbox workload
3
Meta channels normalized into one conversation model
LLM
agent layer trained around conversation patterns and support context
State
Meta event normalization
Webhook events, channel metadata, message direction, sender identity, and account context had to collapse into one reliable conversation model.
Runtime
20K WebSocket fan-out
Django Channels, Daphne/Uvicorn, Redis groups, connection lifecycle, and backpressure behavior were treated as core product infrastructure.
AI
Conversation-aware agents
The agent layer used historical message patterns and active thread context for reply assistance without breaking the operational inbox flow.
Stack