Skip to content
Rohan Patil

Rohan Patil

Senior Software Engineer

Founding engineer on Contentstack Launch. The layers under the deploy button.

I work on the parts of a hosting platform you only notice when they fail — build orchestration, edge routing and auth, cache invalidation, rollbacks, security scanning, and the observability to prove any of it works. On Launch since 2021, across AWS, Google Cloud, Azure and Cloudflare.

Read the case studiesHow the platform is built
Request path through Contentstack LaunchCode from a Git provider flows through build orchestration into immutable artifacts, is served by multi-cloud runtimes behind an edge layer, and reaches the customer domain. A dashed loop shows instant rollback as a routing change back to a previous artifact. Boxes outlined in the accent colour are the parts I designed and built.FIG. 01 — REQUEST PATH, CONTENTSTACK LAUNCHWHAT I DESIGNED AND BUILTGit providerGITHUB · BITBUCKETBuild orchestrationMULTI-STAGE · FAIL-OPEN▲ scan · 300s capArtifactsIMMUTABLE · LAST-KNOWN-GOODRuntimesCONTAINERS · SERVERLESS · STREAMINGAWSGCPAZUREEdgeNGINX/OPENRESTY · CLOUDFLARE WORKERScache-tag & hostname purgeorg SSO for private envs · kill-switchinstant rollback = routing updatecustomer-domain.comANALYTICS → CLICKHOUSE · P99 < 1S← rollback
Fig. 01
years engineering
7+
2018 → present
clouds in production
4
AWS · GCP · Azure · Cloudflare edge
Jira projects contributed to
19
across Launch work, 2021 → now
platform deliveries, end to end
3
FY25‑26: Bitbucket Cloud, IAM keys, cache revalidation

The platform

Launch, and the layers I've owned

A multi-cloud hosting and deployment platform for enterprise websites: Git-connected builds, containerised and serverless runtimes, an edge layer with cache control and streaming functions, and the routing that puts a deployment behind a customer's domain.

Each layer below is a one-line summary; the numbers link to the case studies that go deep.

  1. 01

    Build & delivery plane

    — Multi-stage build orchestration; GitHub, Bitbucket Cloud and plain upload as sources.Study06
  2. 02

    Multi-cloud runtimes

    — Containers and serverless on AWS, Google Cloud and Azure, with HTTP streaming on all three.
  3. 03

    Edge routing & auth

    — NGINX/OpenResty and Cloudflare Workers: cache-tag purge and priming, org SSO for private environments, rollback as a routing change.Studies0204
  4. 04

    Public API

    — A REST layer beside GraphQL, one merged OpenAPI surface, pen-tested before exposure.Study05
  5. 05

    Security

    — IAM roles in place of long-lived keys, Redis on TLS via a live cutover, customer code scanning designed from a blank page.Studies0107
  6. 06

    Observability & analytics

    — Event and log pipelines on Kafka, RabbitMQ, Redis and OpenSearch; HTTP traffic analytics on ClickHouse with sub-second percentiles.Study03

Also shipped

  • HTTP streaming across the clouds

    Chunked transfer encoding on every cloud Launch runs on, a per-environment toggle, basic-auth in streaming mode, benchmarks and load tests after the change landed.

  • Cache revalidation customers can drive

    Purge by domain and by cache tag off the header NGINX sets, wired into the Automate connector so an editor can trigger it without an engineer; a Kafka-backed batching design for volume.

  • Multi-cloud resource cleanup

    Orphaned serverless resources and CloudFormation stacks removed across AWS, Azure and GCP, then a cloud-agnostic cleanup script so it stops being a manual migration.

  • On-call, benchmarks, checklists

    Set up on-call for Launch services, benchmarked deploy time and TTFB against Netlify, Vercel and Gatsby Cloud, and wrote the Go-Live checklist new customers are onboarded with.

Case studies

Seven pieces of Launch, in depth

Each one is a real problem, the design that answered it, the decisions that mattered, and the number that says it worked. Written the way I'd explain it to another engineer.
  1. 01

    Security · new product area

    Customer code security scanning

    A two-plane scanning architecture — SBOMs, vulnerability matching and secret detection over every customer deployment — with a fail-open guarantee so it can never block a deploy.

    • 2planes: build & analysis
    • 300shard cap, fail-open
    • 74%rule overlap found, Trivy vs Gitleaks
  2. 02

    Edge · architecture

    Edge SSO for Private Environments

    A two-stage Cloudflare Worker pipeline that locks a customer's non-production sites to their own organisation's SSO — with session handling that survives a 60-minute token expiry and an instant offboarding kill-switch.

    • 2stage edge pipeline
    • 60 mintoken expiry, decoupled from session
    • 0trailing access after offboarding
  3. 03

    Observability · data engineering

    HTTP traffic analytics on ClickHouse

    From a CDN-analytics spike to a shipped query API: a ClickHouse engine that gives customers a real-time view of requests, bandwidth, status codes and latency percentiles — sub-second, on pre-aggregated data.

    • <1sp50 / p90 / p99 queries
    • 1 minpre-aggregation grain
    • 4telemetry sources unified
  4. 04

    Reliability · design

    Instant Rollbacks

    A technical design for rolling back a bad deploy at the routing layer — rewiring which deployment receives traffic instead of rebuilding — so recovery costs a routing update, not a build.

    • 1routing update per rollback
    • 0rebuilds
  5. 05

    Platform · API

    Making the Launch API public

    A REST layer alongside the internal GraphQL API, a single merged OpenAPI surface, public domains, pen-tested and documented before exposure.

    • 2services given REST layers
    • 1merged OpenAPI spec
    • 1InfoSec pen test before launch
  6. 06

    Integrations · epic

    Bitbucket Cloud, end to end

    Bitbucket Cloud as a first-class Git provider — OAuth, token lifecycle, persistence, backend, UI, a public Marketplace app, and the procurement and vendor review to get there.

    • 5teams coordinated
    • 3environments staged to public
    • 1public Marketplace app
  7. 07

    Security · epic

    Removing long-lived credentials

    Static IAM access keys replaced with IAM roles across the platform's services, and Redis moved to TLS with a live dual-write cutover — the security debt that is easy to defer forever.

    • 0static IAM keys left in Launch services
    • 0downtime for the Redis cutover

How I work

Four habits the case studies keep showing

  1. 01

    Shipped, measured, proven — in that order.

    Streaming didn't count as done when it deployed; it counted when the benchmark report and the load test after the change said the same thing. If a claim can be measured, I'd rather measure it than assert it.

  2. 02

    Decide the expensive question first.

    Whether a scanner can run embedded and what its memory floor costs decides an entire architecture. I answer that before writing product code, so the cheap decisions downstream stay cheap.

  3. 03

    Put invariants in code, not configuration.

    Two access modes that must never coexist are checked at the edge worker on every request. Fail-open has a hard cap in the collector. Configuration drifts; enforced invariants don't.

  4. 04

    Cleanup is part of the design.

    A rollback design that doesn't say when old deployments are deleted is half a design. Orphaned cloud resources bill until someone notices — so the cleanup script ships with the feature.

Rohan Patil

Seven years in, most of them on one platform. Before Contentstack I built admin platforms and back-office systems for enterprise clients — including a site on Contentstack's CMS, as a customer.

I like problems where the answer can be measured, and I'd rather write the design doc before the code than the post-mortem after it.

Languages
GoNode.jsTypeScriptJavaScriptLua
APIs & architecture
RESTGraphQLgRPCEvent-driven systemsDistributed systems
Cloud & infrastructure
AWSAzureGoogle CloudCloudflareKubernetesDockerTerraformCloudFormationGoCD
Data & messaging
KafkaRabbitMQRedisMongoDBElasticsearchClickHouse
Edge & delivery
NGINX/OpenRestyCloudflare WorkersCache control
Frontend
ReactReduxNext.jsReact Testing Library
Practice
Trunk-based developmentTDDObservabilitySecurity reviewsDesign docs

Career

2018 to now, in order

Two companies, and most of the story sits inside the second one. Drawn as a trace: each span is what I was building at the time, and the nested ones are the years of Launch. Open a span for the detail.
Fig. 02 — career, as a traceselect a span

  • Two-plane customer code scanning designed and prototyped; HTTP traffic analytics built on ClickHouse; edge SSO for private environments; the Instant Rollbacks design.

Read the case studies

Education

  • Master of Computer Applications

    MET Mumbai · 2021

  • B.Sc. Information Technology

    Viva College · 2014

Recognition

Awards, and the things that don't fit a ticket

  • 2024

    Above and Beyond Award

    Contentstack

  • 2019

    Rising Star Award

    Raw Engineering

  • 2023

    Set up on-call for Launch services

    Defined the rotation and paging for the platform's services before there was one.

  • 2023

    Go-Live checklist adopted for customer onboarding

    Reliability, security, performance and cost — written once, handed to every new Launch customer since.

  • 2023

    Competitive benchmarks vs Netlify, Vercel, Gatsby Cloud

    Deploy time and TTFB, head to head, used to set platform priorities.

  • FY25‑26

    Three platform deliveries in one fiscal year

    Bitbucket Cloud integration, IAM key removal, cache revalidation — each owned end to end.

  • 2025

    Vendor risk assessment & procurement for a Git provider

    Unusual scope for an engineer: the paperwork that let the Bitbucket integration exist at all.

  • 2025

    Public API pen-tested before exposure

    InfoSec review as a gate, not a follow-up.

Writing

Notes from the platform

Design decisions, incident write-ups and the occasional strong opinion about cache headers. Nothing here yet — the pipeline is ready when the first one is.

Status

No posts published yet. The first one is likely to be about why a security scanner should never be the reason a deploy fails.

Writing index

Contact

Talk to me about platforms that have to stay up.

Distributed systems, deploy infrastructure, edge routing, security reviews, or a design doc you want a second pair of eyes on. Email is fastest; I reply from Mumbai, IST (UTC+5:30).

rohan.p@outlook.inRésumé (PDF)