All case studies
FinTech

FinTech Trading Platform

A high-frequency trading platform clearing over a million transactions a day, rebuilt around a sub-millisecond matching core - migrated live, with zero trading-hours downtime.

Industry
Capital Markets
Year
2024
Timeline
28 weeks
Team
7 engineers
01 · Challenge

A trading engine hitting its ceiling

A European brokerage came to us with a decade-old monolith buckling under its own growth. Order round-trips averaged 380ms, market-open spikes triggered weekly incidents, and compliance reports were still assembled by hand every night.

  • 380ms median order round-trip
  • Weekly incidents at market open
  • Manual end-of-day compliance reporting
02 · Approach

Event-driven core, boring edges

We isolated the hot path and rebuilt it as an event-driven core: a lean order gateway, an in-memory matching engine fed by Redis Streams, and CQRS read models in PostgreSQL for everything that can afford to be a few milliseconds behind.

Order Gateway
Node.js · auth in under 5ms
Matching Core
In-memory · Redis Streams
Ledger & Reporting
PostgreSQL · CQRS read models
03 · Build

A live migration, six releases deep

Over 28 weeks we ran a strangler-fig migration: dual-writes into the new core, shadow traffic replayed against production data, and six incremental cutovers - all without a minute of trading-hours downtime.

  • Dual-write bridge with automated diffing
  • Shadow replay of full production order flow
  • Zero trading-hours downtime across six cutovers
04 · Results

Numbers on the tape

The new core makes matching decisions in under a millisecond; a full order round-trip, wire to wire, now lands at 12ms. The platform clears more than a million transactions a day with headroom to spare.

0ms
Median order round-trip
0×
Faster than the legacy stack
0.99%
Trading-hours uptime
0M+
Transactions per day
05 · Client

From the trading desk

What the client took away from the engagement, in their own words.

D1VERSY rebuilt our engine while the market was open. Nobody on the desk noticed anything - until the latency graphs fell off a cliff.
Markus R. · Head of Trading Technology