_

Building the gaming platform from zero to ~700K req/min.

joinjumbo.com · New Delhi · Aug 2024 — Present

• Scaled the gaming platform from ~15K to ~700K req/min for ~100K+ MAU (80K+ weekly active), reliably handling ₹5 Cr peak payment throughput and 30 Cr+ in-game transaction volume.
• Rewrote the core transaction service in Go + gRPC, cutting latency from ~150 ms to ~50 ms; engineered financial correctness via atomicity, idempotency, running balances, and integer currency normalization to eliminate float drift.
• Owned end-to-end system architecture for the in-house cricket fantasy game (like Dream11) — service boundaries, schema, and API contracts.
• Designed a multi-currency wallet ledger with liability tracking and automated clawback reconciliation for accurate, auditable balances across product lines.
• Built the fantasy contest engine — concurrency-safe contest join, automated reward calculation & distribution, live leaderboards, and refund / cancellation flows.
• Built a vendor-agnostic event pipeline: one typed event is batched, streamed via Redis, and fanned out to multiple analytics vendors (CleverTap, PostHog, Matomo) while persisted first-party in ClickHouse — removing third-party lock-in.
• Built a CDC pipeline using Postgres logical replication to stream production data into BigQuery in near real time, without loading the primary database.
• Delivered a scalable WebSocket platform (API Gateway + Redis) for connection state, channel broadcasts, and WS-based API calls.
• Rolled out Datadog observability (APM, metrics, dashboards, alerting) across all production services, improving incident detection and response.
• Primary escalation owner for production incidents — debugging, resolution, release execution, and unblocking the team.
• Built an async recharge + KYC orchestration layer with a vendor-agnostic Adapter Pattern, reducing third-party cost by ~70%.
• Cut backend load and CDN cost by ~40–50% by moving low-change / high-read APIs behind a CDN with correct cache-control and invalidation.
• Scaled database reads with read replicas and managed high concurrent connection volume via RDS Proxy + pgBouncer pooling, keeping the primary stable under peak load.
• Reclaimed ~30% of database storage with pg_repack (table rebuilds, reindex, dead-tuple cleanup) and upgraded PostgreSQL 17 → 18.
• Improved tail latency and peak resilience through API + DB bottleneck analysis and load testing with Locust and Percona.
• Eliminated OOM crashes in the Go transaction service by setting GOMEMLIMIT — the GC runs more aggressively as the heap nears the cap, keeping memory stable under load.
• Built centralized logging + trace correlation on Node.js async_hooks, reducing MTTR and surfacing perf bottlenecks across services.
• Modernized AWS infra with AWS Organizations, CI/CD pipelines, IAM Identity Center RBAC, and Tailscale-secured internal access.
• Performed zero-downtime rotation of JWT signing keys (kid-based) across all production services, with no forced user logout.
• Strengthened account safety with OTP-abuse prevention and multi-device concurrency conflict handling.
• Built an Aviator game from scratch, scaling 0 → ~1,000 CCU; improved velocity using AI-assisted tooling and MCP database workflows.