Dhruv Agrawal Software Engineer MyRik · Bengaluru
DhruvAgrawal
I build the systems under the surface.
Backend engineer. The platforms I run serve 150,000 people a month; their latency, cost and uptime are my job. Every claim on this page is backed by something you can run. Scroll to go one layer down at a time.
01 People 150,000 / month
The people on the other end of the API.
Users never see a backend. They feel it as a ride that books instantly, a page that loads under 300 ms, a video that plays. These are the numbers behind that feeling, each one from a system I owned.
Before MyRik: CarWale (CarTrade Tech) and JioHotstar. B.Tech in Computer Science, LNCT Bhopal, 8.23 CGPA. 800+ DSA problems, LeetCode Knight 1870+.
02 Systems 7 services
Systems I've run in production.
On the right: the real topology of the real-time collaboration platform, Spring Cloud Gateway as the only public port, Eureka holding the address book, packets moving the way they do in the code. On the left: where I've done this for a living.
- Zero-downtime GCP → AWS migration with a hybrid VPC-tunnel phase; observability re-platformed to self-hosted Grafana / Loki / Tempo.
- p99 from >1 s to <300 ms (OpenTelemetry, N+1, indexes, cross-region RTDB).
- CQRS, event-driven product service for 150K+ MAU; Razorpay flow with webhook reconciliation.
- Desktop Short Videos and AI chat support for 1M+ monthly users: +15 % engagement, 50 % faster query resolution.
- 20 % fewer Kubernetes pods, 25 % lower infra cost; legacy frontend to 70 % coverage, 40 % fewer defects.
- Kafka → GCP Pub/Sub with pull-based consumption: 5M+ content-quality events a day at 99.9 %; Go APIs for 10,000+ jobs a day.
03 Machines 3 runnable
Machines you can run.
Three projects, each with a working demo instead of a screenshot. On the right: DStarDB's reactor loop feeding a worker pool, the shape that beats single-threaded Redis under concurrent reads.
Redis-compatible in-memory store. Event loop reads bytes, N workers execute, reader-writer concurrency; 10,000+ concurrent connections; YCSB: 13 % more throughput, 50 % lower latency than Redis under concurrent reads. The REPL runs the browser build in a Web Worker; every reply is timed in microseconds. Case study →
Seven services, one capability each. Signalling carries SDP and ICE over one topic per room; media goes peer to peer and never touches the servers. The case study opens a real RTCPeerConnection: create a room, share the link, join from any device.
PIB press releases in, narrated multilingual videos out. Cut production time by 45 %.
Twenty-four services, real queues, real timeouts. Kill a worker, partition the cache, slow a dependency, then watch circuit breakers, failover and the autoscaler put it back. Replay the incidents I actually fixed, step by step.
Also shipped: Warehouse CCTV anomaly detection (35 cameras, 8 warehouses, MyRik) · College attendance app (Flutter + Node, 6.5K downloads) · Amazon and WhatsApp clones (Flutter, Firebase, Riverpod).
04 Stack 6 layers · 36 tools
The stack, by layer.
Not a wall of logos. A cross-section of the tools I reach for, ordered the way a request passes through them. Hover a layer to light it up.