TOKI
Mongolian Super App (Payments & Multi-Services)
Large-scale super app supporting digital payments and multiple consumer services. Backend served 500K+ MAU with real-time location and notifications.
Impact — P95 API latency under 200ms at peak; 99.9% uptime; zero-downtime deployments and backward-compatible API evolution.
Problem
Consumer payments and multi-service flows needed sub-200ms critical APIs, real-time location triggers, and releases that never took the app offline for hundreds of thousands of monthly users.
Approach
Event-driven Node.js services on AWS with Kafka for async workflows, MongoDB for core data, ScyllaDB for hot location/notification paths, and strict backward-compatible API evolution across client versions.
What shipped
- Designed and built Node.js backend services serving 500K+ monthly active users
- Zero-downtime deployment strategy; rolled out releases with no service interruption for end users
- Backward-compatible API and contract evolution; supported multiple client versions without breaking existing consumers
- Event-driven workflows using Kafka; reduced notification delivery time by ~60% vs batch jobs
- MongoDB for core application data; ScyllaDB for real-time user location and notification triggers at scale
- AWS integrations (API Gateway, Secrets Manager, S3, SNS, SES); cut cloud cost ~15% via query and cache tuning
- Achieved 99.9% uptime and sub-200ms P95 latency for critical payment and location APIs
Stack
Related writing
- Why Event-Driven Architecture Matters for Scalable Backends
How Kafka-style event pipelines reduce coupling, improve resilience, and help teams ship faster without breaking production.
- Zero-Downtime Deployments with Node.js
Rolling updates, health checks, and backward-compatible releases so users never see a 502 during deploy.
- Backward-Compatible API Versioning in Production
How to evolve REST and gRPC APIs without breaking mobile clients and partner integrations.