Writing & Thinking
Thoughts on engineering leadership, AI tooling, and building platform teams at scale.
Building an Agentic CI/CD Pipeline: Four Agents, One Human Gate, and What Actually Breaks
Engineering attention is the scarcest resource in any team. ai-sdlc is a four-agent pipeline — BUILDER, TESTER, REVIEWER, CHECKER — that takes a GitHub issue and delivers a reviewed, gated PR in under 15 minutes, for under $4 in AI compute. This is how it works.
Building Realtime Messaging Infrastructure for 50M+ Concurrent Users (Part 1)
How we built Hotstar's PubSub broadcast infrastructure for 50M+ concurrent users during IPL 2019 — protocol selection, EMQX clusters, load balancing, and the reverse bridge pattern that achieved sub-second latency at scale.
Building Realtime Messaging Infrastructure for 50M+ Concurrent Users (Part 2)
Extending Hotstar's PubSub system for unicast and multicast delivery to 50M concurrent users — the O(1) service discovery approach, sub-second latency results, and the product features it unlocked.
Capturing A Billion Emo(j)i-ons
How Hotstar processed 5 billion emojis from 55M users during the 2019 Cricket World Cup — Kafka, Spark Streaming, and an in-house system that scaled to power Voting, Polls, and Trivia for major Indian reality shows.
Why I Enjoyed Working at Disney+
From shipping Watch'N Play as a consultant during IPL 2018 to co-building a PubSub system that processed 700 billion messages in total — what I learned in 18 months at Disney+ Hotstar and why it shaped how I think about engineering culture.
Scheduling Tasks in Android Made Easy
Picking the right Android scheduling API across versions — AlarmManager, JobScheduler, GcmNetworkManager — and why we built Smart Scheduler at HyperTrack.