Rebuilding a trading platform’s execution layer under live load
A re-architecture of a FinTech execution engine from a single monolith into an event-driven system, carried out without a maintenance window.
- Deploy frequency
- 1×/quarter → 1×/day
- P99 order latency
- −63%
- Execution-path incidents
- 0 during migration
- Timeline
- 7-month engagement
The problem
A mid-stage FinTech’s order execution service was built as a single monolith that could no longer be deployed safely — every release risked a multi-minute outage during market hours, and the team had stopped shipping features to avoid the risk.
Our approach
We mapped the monolith’s actual data flows, identified the three bounded contexts hiding inside it, and extracted them one at a time behind a strangler-fig proxy, each cutover validated against shadow traffic before the old path was retired.
The outcome
Deployments moved from a scheduled, all-hands event to a routine, engineer-triggered action — with zero execution-path outages across the migration window.
Stack
- TypeScript
- Node.js
- Kafka
- PostgreSQL
- AWS
- Terraform