New Delhi, India · Best Performer @ Octro · Open to Backend Roles
Backend engineer building distributed Java and Go systems in production. On-call for 54 services. Exploring LLMs & AI systems.
// System Designs — interactive
Explorable architecture case studies — trace a request through the system, break it, and see the tradeoff behind every decision.
Decoupling push delivery from the request path so a slow third party can never stall the app — how I built it, the tradeoffs, and how it scales to 100M users.
Explore →Merging a guest account into a registered one without losing progress — safe under partial failure, idempotent under retries, auditable end to end.
Explore →Live ranks for a multiplayer game — O(log N) updates, one push per interval instead of per score change, and recovery when the cache dies.
Explore →// Production Work
Designed and shipped a multi-provider identity linking system (Facebook, Apple Sign-In) so users can migrate existing accounts to social login without data loss. Owns the full flow: server-side time-bound token generation, email verification, and an atomic merge transaction across 3 microservices — with guards for concurrent same-user requests, partial-failure rollback, and TTL-based token invalidation.
Built a reward system where each reward carries its own independent TTL — expiry is per-reward, not per-session, so users keep what they've earned across logouts. Integrated with an external CRM for scheduled delivery triggers. Designed the cross-service data contract and maintained reward balance integrity across account merge and deletion flows spanning 3 services.
Re-architected mobile push notification delivery from a blocking HTTP call on the login critical path to a fully async SQS pipeline — decoupling notification throughput from login latency. Built iOS/Android platform routing, per-platform delivery rate limits backed by ZooKeeper runtime config, and a dead-letter queue with CloudWatch alerting for failed deliveries.
Built a compute layer bridging the Java microservices and versioned TypeScript AWS Lambda functions — Java resolves the correct Lambda version from ZooKeeper runtime config and invokes it asynchronously via CompletableFuture, with the computation logic itself authored in TypeScript. Recognized as Best Performer for this and related backend work.
Built the backend for a real-time multiplayer session serving concurrent player connections — concurrent state management, per-session configuration, and a live leaderboard broadcast to all connected clients over WebSocket every 30 seconds. Maintained per-session state isolation under concurrent load.
Made services configurable through the platform's ZooKeeper-backed runtime config, propagating feature flags and operational parameters to all 54 service instances within seconds, with no redeployment. Removed the deploy-to-change cycle for operational settings — enabling zero-downtime staged rollouts, kill-switches, and per-environment overrides.
Diagnosed a production incident where IllegalMonitorStateException on distributed lock release was causing 5xx bursts across 3 services. Root cause: lock acquired on thread A, unlock attempted on thread B after an async context switch — violating the lock library's thread-affinity contract. Fixed the unlock call-site and added thread-ID correlation to distributed tracing.
Resolved two concurrent production incidents: (1) a race in WebSocket reconnect handling where non-atomic session state transitions caused duplicate authorization and silent message drops; (2) a NullPointerException in the reward delivery pipeline that blocked reward delivery for all active users until patched.
Identified and fixed two N+1 patterns: a polling service issuing individual Redis round-trips per user on every cycle (replaced with a single batched read), and MongoDB fan-out on the login critical path (batched into one query). Also corrected Maven dependency scoping, halving the deployed artifact size.
// Side Projects
LeetCode-style backend in Go — worker pool with buffered channels, ephemeral Docker containers per submission with resource limits and context timeouts, SSE streaming for real-time output.
View on GitHub →Production-style Go load balancer with multiple routing strategies, runtime config reload without dropping traffic via fsnotify + sync.RWMutex, and active health checks with failure isolation.
View on GitHub →// Stack
// Contact
Open to backend engineering roles in distributed systems, platform engineering, or AI/LLM infrastructure.