p99 4ms · 300K rps · all green

Sarmad Shah

Frontend-focused full-stack engineer — 8+ years shipping pixel-perfect, real-time UI for millions at Coinbase and Zomato, on top of tier-zero infrastructure I ran myself.

The hardest problems live at the seams between layers.

300K
RPS peak, tier-zero
99.999%
availability SLO
$640K+
infra saved / year
15→2 min
incident diagnosis, via AI agent
10K+
bounty submissions
Sarmad Shah

3-minute live session · AI avatar

Talk to me — right now.

A photoreal AI avatar of me, trained on my work. Ask it anything a first call would cover.

About

The seams between layers

Most engineers pick a layer and stay there. I've spent eight years refusing to — translating designs into pixel-perfect, animated interfaces used by millions at Coinbase and Zomato, while serving as DRI for a tier-zero config service holding 99.999% availability at 300K RPS.

That range isn't trivia. The hardest production problems — a real-time dashboard dropping frames under a firehose of updates, a 2,300-secret migration where any failure is an org-wide SEV-1 — live at the seams between frontend, backend, and infrastructure. Owning both sides is how you fix them.

Lately that includes AI: an incident-response agent that cut diagnosis from 15 minutes to 2, and deep frontend work on streaming, token-batched AI interfaces — the kind powering the chat on this page.

Work · 2018 → 2026

Eight years, four chapters

Deep-dive cards expand. Everything below is shipped, production work.

Coinbase

Senior Software Engineer · Tech Lead — Jan 2022 – May 2026 · Gurgaon

Core Infrastructure — Reliability · Frontend Tech Lead · 2023–2026

deep dive

Config Service SDK

A TypeScript SDK adopted by every Next.js app at Coinbase — coinbase.com and Coinbase NFT included — wrapping the tier-zero config service behind a clean, type-safe API for config and secret access.

every Next.js appAES-256-GCMS3 failover
  • Designed for the Next.js server runtime — secrets never reach the client bundle.
  • Caching layer for fast SSR, exponential backoff, and S3 failover when the service is unreachable.
  • Envelope encryption with AWS KMS + AES-256-GCM — data keys decrypted via scoped, temporary KMS grants (least privilege), secrets decrypted locally.
  • Secured secret delivery for Coinbase's entire Next.js frontend surface — org-wide leverage from one SDK.
deep dive · flagship

Scale-In

Services safely stress-test themselves in production to find the CPU ceiling they can run at — then run fewer pods, hotter. Led company-wide, end to end, across both halves of the stack.

300+ servicesCPU 30%→50%~$300K/yr

backend

  • Ephemeral K8s pipelines orchestrated by Temporal workflows.
  • Pod cloning — ALB → target group → service → pods, identical runtime identity, same namespace.
  • Small % of real traffic shifted; HPA disabled on the clone so CPU actually rises.
  • Traffic-ramp control loop with guaranteed teardown.

frontend

  • Live CPU/traffic charts + streaming activity log with a stop control.
  • Two-tier WebSocket subscriptions multiplexed over one connection; snapshot-then-subscribe.
  • useSyncExternalStore + selectors — each panel re-renders only on its own slice; races keyed by execution id.
  • rAF-batched charts, virtualized log, LDAP/Git-driven RBAC.

AI incident-response agent

LLM reasoning with tool-calling over observability data — Datadog logs, deploys, anomalous call patterns, AWS/UI health. Diagnosis 15 min → 2 min. Plus a test-automation CLI orchestrating e2e end to end.

Tier-zero infrastructure range

DRI for the T0 config service — 99.999% SLO at 300K RPS, scaled 3,000 → 16,000 dependent services. Zero-downtime Lambda → Kubernetes migration with AWS IAM → SPIFFE/SPIRE auth ($40K/mo saved); Redis → Valkey (3× throughput, −70% latency); 2,300 production secrets moved across AWS accounts without a blink.

Config Service UI, V1 → V2

Rebuilt the console for a T0 service used by 16,000+ services — batch updates, an inline proto-aware editor with real-time schema validation. Page load 9s → 2s, 300% faster navigation.

Bounty platform · Frontend Tech Lead

Built the frontend for Coinbase's developer bounty program (solve bounties → earn interviews) with Coinbase Wallet auth — 10,000+ submissions. As FE Tech Lead across India + US: org-wide patterns, mentorship, and an NX monorepo migration cutting per-change PRs 4 → 1 for ~30 engineer touchpoints a month.

International — Frontend · 2022–2023

deep dive

Server-Driven UI framework

One backend-driven schema rendering natively on web and React Native. Replaced 100% of static app pages and eliminated frontend deployments entirely.

web + RN100% of static pageszero FE deploys
  • A cross-platform contract: component registry, stateful components, dynamic action handling.
  • Full Coinbase Design System support — designs translated into pixel-perfect components on both platforms.
  • Instant UI updates with no app-store release — the backend ships the interface.

KYC & onboarding — India + Japan launches

End-to-end flows for two market launches — funnel drop-off analysis, measurable completion-rate gains, country-specific regulatory compliance.

Rewards flow + spin wheel

The animation-craft showcase — gesture-driven motion, easing, haptic feedback. Fluid micro-interactions that drove engagement.

Zomato

Software Development Engineer II — Aug 2020 – Jan 2022 · New Delhi

deep dive

Support platform + Chat SDK

Led a team of 4 to build Zomato's entire customer-support platform from scratch, replacing a legacy Firebase stack — on top of an MQTT Chat SDK that outlived the project.

5× parallel queries−50% op cost10K+ tickets/day
  • Real-time Agent Dashboard: 200+ simultaneous WebSocket connections with complex state sync and zero lag.
  • The war story: started on React Context — every conversation re-rendered on every message. Measured it, moved to Redux + selector subscriptions. Changed my mind with data.
  • Zomato Chat SDK (MQTT) as the real-time backbone — adopted across Zomato products and by Blinkit post-acquisition; integrated chatbot for automated flows.

TypeScript pioneer · in-app WebViews

Introduced TypeScript across a previously vanilla-JS frontend codebase; component standards and design-system patterns adopted org-wide. Built dynamic WebView pages rendered inside the Zomato app for urgent banners and flows.

Crownstack

Frontend Lead — May 2019 – Jan 2020 · New Delhi

Velostics

Led frontend end to end — driver mobile app (real-time location tracking, offline-first, multi-stop routing) and web platform (RBAC, complex dependent forms, live driver tracking).

Miracle Studios

React Native Engineer — Sep 2018 – Mar 2019 · Chandigarh

Jerry

Sole React Native engineer in the org — built a SaaS mobile product from scratch; sold to multiple clients.

Skills

The range, itemized

Frontend

React · React Native · Next.js · TypeScript · Relay · TanStack Query · Zustand · Tailwind · Webpack · Vite

Design & UX

Design systems · Pixel-perfect UI · Micro-interactions · Framer Motion · Reanimated · Accessibility

Backend & Infra

Go · Node.js · GraphQL · Kubernetes · AWS Lambda · AWS KMS · SPIFFE/SPIRE · Temporal · Redis · Valkey · DynamoDB · Datadog

AI

Claude (Anthropic) · Cursor · AI-powered CLI tooling · Custom LLM plugins & skills · Prompt engineering

Real-time

WebSockets · SSE · MQTT · pub/sub architectures · streaming UIs

AI work

Agents that act, UIs that stream

Incident-response agent

LLM-powered reasoning with tool-calling over live observability data: it pulls Datadog logs, inspects deploys, flags anomalous service-calling patterns, and checks AWS/UI health — structured outputs driving real diagnostic actions during incidents.

diagnosis: 15 min → 2 min

Streaming UIs for AI products

The frontend side of AI: SSE and fetch + ReadableStream token rendering, rAF token batching, streaming markdown, RAG with citations — real-time interfaces that stay smooth under a firehose of updates.

the chat widget above is the demo

Open source

react-native-easy-content-loader

Skeleton content loaders for React Native — the loading states you don't have to hand-build.

github.com/sarmad1995

Off the clock

Calm under pressure, on purpose

Skiing — 2× national competitor. Brazilian Jiu-Jitsu — blue belt. Swimming, chess, esports. Sports where you find out fast whether you're actually good or just think you are — the same nerve that keeps a 3am SEV-1 boring.

Contact

Let's talk — for real this time.

The avatar covers the first three minutes. Everything after that is me.

Email meGitHubLinkedInResume

sarmadshah.work@gmail.com · +91-7006576673 · India

© 2026 Sarmad Shah — this page is a live demo of the craft it describes.