Senior Backend Engineer · Distributed Systems & Platform Infrastructure · Python · AWS
I build the shared infrastructure that 200+ production APIs run on. I work at Perfios, where I scale and maintain the core APIs of TotalKYC, an identity-verification platform.
- Cost: cut third-party vendor spend 55% by rebuilding the routing and caching layer; with retired infrastructure, cumulative savings reached $700K+ since Apr 2024.
- Reliability: designed and set up active-passive disaster recovery across Mumbai and Hyderabad for a platform serving 5M+ daily requests at 99.9%+ availability, built for a 40–45 minute region switchover.
- Platform: designed and rolled out shared-utilities (secrets access, schema validation, request configuration) across 200+ production APIs. A fleet-wide change now ships as one package update instead of a redeploy of every API.
- Routing: built health-aware routing with cooldowns for 120+ APIs, replacing random route selection and raising first-attempt success from 60–70% to 90–95%.
- Coordination: built a Redis-backed session library with distributed locks for concurrent Lambda runs, cutting session failures by 35% and recovering 15K–30K requests a day.
- Operations: brought incidents from 12–13 a week to 2–3 with trend-based monitoring.
Where they show up:
- Python: 260 APIs moved from 3.9 to 3.12, zero rollbacks
- Redis: session locks; the quota-meter Lua script
- Celery + ECS / Fargate: long-running workers, 3× throughput
- AWS Lambda: concurrent runs kept from racing by the session library
- Docker: 245 APIs containerized
- GitLab CI/CD: the DR switchover pipeline
Also in the toolbox: FastAPI, Node.js, JavaScript, SQL, C++, MongoDB, MySQL, RabbitMQ, Amazon SQS, S3, CloudFormation, AWS CloudWatch, Prometheus, Grafana, REST APIs, Microservices, Event-driven architecture, Pytest, Playwright, Selenium, Security
A per-organization, per-feature quota service. A single Redis Lua script validates, deduplicates, deducts, and records denials atomically, with all-or-nothing batches, idempotent refunds, and cron-free monthly resets. Load-tested 1,000 requests across 50 concurrent workers and 3 service instances with zero over-serving; same-key retries are charged once.


