Starting point
Members see their next and historical payments on My Page. The solution was a Java monolith that loaded slowly and produced errors, because the domain rules and the members’ special cases were missing from the code. The settlement team works with postings, not with what the member should see, so periodization and consolidation of amounts, lines per product and back payments had to be built in our solution.
What I did
- Rewrote the entire logic in Kotlin/Spring Boot, with the rules built together with the settlement experts.
- Verified the new code against the old one in production: a comparison logged every discrepancy, each one was investigated, and in the end the remaining discrepancies were bugs in the old code.
- Built a pipeline for bug fixing: fetches source data from settlement, anonymizes and filters it down to the failing case, and gives an AI agent what it needs to find the bug and open a PR, which then goes to functional testing. No sensitive data reaches the AI agent.
- Architect for moving logic from the monolith to microservices, in parallel with the team’s regular deliveries. Rewrote a Node/TypeScript application in Kotlin.
- Two initiatives demonstrating frontend and backend in GCP against on-premise services, one together with Google.
Result
Response time improved four times over, and the last discrepancies against the old code turned out to be bugs in the old code. The result led to the role as architect and backend developer for SPK’s new pension calculator.