The platform was a partner portal for a large fintech company, used by more than 100,000 active users across seven partner role types. Underneath it sat a legacy .NET monolith with roughly 1,200 endpoints. Releases took over six hours, every change felt risky, and scaling one hot path meant scaling everything.
- ~1,144 endpoints now running across 3 Go services on PostgreSQL and AWS
- Deploy time cut from 6h+ to under 30 minutes
- ~99.997% uptime, about 15 minutes of downtime a year
- Delivered over 14 months
Few services, clear boundaries
"Microservices" doesn't mean hundreds of tiny services. Three Go services with clear ownership boundaries were enough to unlock independent deploys and scaling without adding a distributed-systems tax the team didn't need.
Load-test against reality, not hope
Before cutting over, we built a dedicated performance-test environment: a read replica of production data plus a custom load generator running at 4× peak traffic. If the new path held at four times the worst day, cutover stopped being scary.
Security as a migration requirement
A migration is the cheapest moment to fix authorization. In the new services, every API call checks that the client actually maps to the partner making the request (closing off IDOR-style data leaks). Secrets live in AWS Secrets Manager rather than in config or code, and AWS access is role-based with no static credentials.
Faster deploys came from process, not hardware
The drop from six-hour deploys to under thirty minutes didn't come from a faster CI box. It came from a new branching strategy, a proper release-management flow, and automated GitHub Actions workflows.
The old flow was the real bottleneck: before a release, unfinished work had to be removed from the develop branch and only then merged into release, a step that alone took about 1.5-2 days. Now changes merge into a release branch and the release merges to master, with the pipeline doing the repetitive work. Nothing has to be pulled back out, so releases stopped being an event.
Shared services need a front door too
Alongside the migration, shared sub-services used across multiple UIs (Portfolio, Tagging, Ledger, Profile and others) had their service-to-service traffic moved onto Kong API Gateway. Putting that traffic behind one gateway gives consistent routing, auth, and visibility instead of every consumer wiring things up its own way.
Operate it like it matters
Every P0 incident gets a root-cause analysis with a two-hour resolution target. That discipline, more than any single technology choice, is what keeps uptime near 99.997%.