Backend Engineer · Hyderabad, India
I build backend systems that stay correct when things go wrong.
Since 2024 I have built the services at Khumbu Systems that connect delivery and ordering apps to the point-of-sale systems of about 12,000 restaurants. On the side I built a real-time multiplayer chess platform to go deeper on concurrency and failure, then spent as long trying to break it as building it.
- ~12k
- restaurant stores in 17 markets served by services I own
- ~75%
- lower cache cost after leading a move to ElastiCache Serverless
- 100/100
- rounds with exactly one winner, 16 requests racing for the same move
- 40 / 0
- live games carried through a rolling deploy / games lost
Selected work
Side project · 2026
Real-time multiplayer chess platform
A server-authoritative clock that cannot drift, exactly-once moves without a distributed lock, and games that survive the server they are on being replaced mid-move. Deployed to AWS with Terraform, run on Kubernetes, load-tested and failure-drilled, with every number written up.
- Java 25
- Spring Boot 4
- PostgreSQL
- Valkey
- WebSockets
- SQS
- ECS Fargate
- Kubernetes
- k6
Production · Khumbu Systems
Redis + Lettuce to Valkey Serverless + GLIDE
Leading the migration of a provisioned Redis cache to ElastiCache Serverless: a custom Spring cache on a client Spring does not support, a shutdown-ordering bug that only appeared during deploys, and the gzip change that cut ECPU usage by 44%.
- ElastiCache Serverless
- Valkey GLIDE
- Spring Cache
- DynamoDB Streams
- SQS
Experience
Khumbu Systems
Associate Software Engineer, Backend
2024 – present
- Design and own the order, cancellation, pricing, menu-sync and store-sync services that connect Uber Eats and RBI’s apps to the point-of-sale systems of Burger King, Popeyes, Tim Hortons, Chipotle and Wingstop.
- Daily: 60k quotes, 10k orders, 40k menu syncs and 123k store-status syncs, peaking at 100 requests per second.
- Idempotent SQS consumers keyed on correlation ID and queue name absorb partner retries and redeliveries; health checks across Valkey, SQS, S3 and DynamoDB drive disaster-recovery failover between regions.
- Cut heavy menu processing from over 70 s to under 20 s; upgraded every backend service to Spring Boot 3 in one week with zero rollbacks; automated rollouts that onboarded 10,000+ stores across EMEA, APAC and the US.
Writing
Medium · 22 June 2026
From Redis + Lettuce to Valkey GLIDE: Lessons from a Production Migration to ElastiCache Serverless
How we reduced cache infrastructure costs, navigated Spring lifecycle challenges, and learned what really happens behind a Serverless cache endpoint.
Read on MediumContact
I enjoy talking about backend and distributed systems, and the trade-offs behind the work on this site. I’m always open to a good conversation, and email reaches me fastest: