Problem
Meridian's settlement engine ran on a 12-year-old batch architecture that couldn't support same-day currency clearing, and every schema change risked a multi-week regression cycle. Engineering velocity had slowed to a crawl exactly when the business needed to move faster.
Strategy
Rather than a rewrite-and-cutover, we re-architected the core around an event-sourced ledger and migrated live traffic in stages behind a shadow-write layer — validating every transaction against the legacy system in parallel before it was ever trusted alone.
The new core centers on an append-only ledger that every downstream system reads from, replacing point-to-point integrations with a single source of truth. A reconciliation engine runs continuously against the event stream instead of overnight, and reporting is derived rather than separately maintained.
Solution
We rebuilt the reconciliation engine to run continuously instead of overnight, replaced brittle point-to-point integrations with an event bus, and gave the team a deployment pipeline that made shipping ledger changes routine instead of terrifying.
Impact
Settlement latency dropped from hours to under four seconds, and the engineering team now ships ledger changes weekly instead of quarterly — with zero downtime across the entire migration.
Settlement latency
Deploy frequency
Migration downtime