A reconciliation API that finance teams trust

The challenge

Ledgerline's customers reconciled payments with a batch job that ran overnight. When it failed, finance teams started the day with stale numbers and a support queue full of "where is my payment?" tickets.

The job had grown over five years into a single 4,000-line script. Nobody wanted to touch it, and every bank added a new edge case.

The approach

• Mapped every bank feed format and failure mode with the support team before writing any code.
• Split ingestion, matching and posting into small Go services connected by Kafka topics, each with an idempotency key so retries never double-post.
• Kept PostgreSQL as the single source of truth and used an outbox table instead of distributed transactions.
• Ran the new pipeline in shadow mode for six weeks, comparing results with the old job every night before switching over.

The result

The overnight job was switched off in a single release with no customer-facing incident. The in-house team now owns the service, with runbooks and dashboards written alongside the code.

“Mira didn't just rebuild the system, she left us a team that understands it. The handover was the best I've seen from a contractor.”

— Head of Engineering, Ledgerline