
Taraka R
Software Engineer Java Backend Distributed Systems Fintech
Kompetenzen

Meine Dienstleistungen

Arbeitserfahrung
SOFTWARE DEVELOPMENT ENGINEER - 2
Zetasoft • Vollzeit
Dec 2024 - Present • 1 yr 9 mos
FD-backed credit cards for India's underserved credit segment. I own the onboarding, KYC and payments backend. Own the application-journey state machine that moves a customer from sign-up to card issuance across 15+ steps. Authored the HLD/LLD, drove the API contracts, and shipped it to production. Resolved a P0 database hotspot on the high-write user _journey_ event table that was blocking product launches. Separated the transactional and reporting paths — moved to a partitioned schema and pushed analytics consumers onto SNS→SQS fan-out instead of reading off the write path. Cut average write latency ~60%. Built the Autopay/e-Mandate payments pipeline (NDPS, Razorpay) for one-time charges, mandate setup and recurring debits. Transactional outbox plus idempotency keys guarantee a charge is applied exactly once under retry storms, duplicate webhooks, or partial failure between our DB and the gateway. Integrated IOB core-banking APIs (CIF and card lifecycle) over OAuth 2.0 with AES/RSA-encrypted payloads, and built a centralized retry framework (exponential backoff + idempotency keys) that recovered ~30% of previously failed applications with zero manual intervention. Reduced p95 latency on critical onboarding endpoints from 800ms to ~250ms (~70%) through composite indexing, selective denormalization, MyBatis batch operations, and Redis caching on hot read paths. Built the VKYC pipeline with webhook callbacks, asymmetric encryption, Resilience4j time-based circuit breakers, and ShedLock-coordinated refund schedulers so jobs fire exactly once across a multi-pod deployment — eliminated dropped refunds. Run the services on EKS: deployment manifests, resource limits and autoscaling, rolling releases, and pod-level debugging during incidents. Tuned Datadog APM with error.expected span tagging for expected business outcomes — cut the false-positive error rate ~40%, stabilized SLO burn-rate alerts, and reduced on-call paging. Drive code reviews, LLD walkthroughs and incident