Skip to content
SASalil Agrawalsite reliability engineer

Site Reliability Engineer · Astronomer

Hi, I'm Salil —

Ikeepproductionboring.

4+ years across AWS, GCP and Azure — building observability that pages the right human, network architectures that don't surprise anyone, and fixing the root cause even when it lives upstream.

Book a callRésumé
  • 2 PRs merged · apache/airflow
  • AWS · GCP · Azure
  • Founder · Uprevol
  • 5k+ incidents closed
  • HackerOne · er_salil
  • Everything as code
  • 4+ yrs on call
lower MTTR
0%
lower MTTR
Alert redesign · Kibana · Robusta playbooks
less alert fatigue
0%
less alert fatigue
Chronosphere monitors as Terraform
customer deployments
0+
customer deployments
Enterprise Airflow on AWS · GCP · Azure
incidents resolved
0k+
incidents resolved
Across 4+ years on call

Tools I run in production: Kubernetes, Terraform, OpenTelemetry, Prometheus, Grafana, AWS, Google Cloud, Microsoft Azure, Elastic, Kibana, Dynatrace, Python, FastAPI, Linux

01Experience

Career, traced.

Four roles on one time axis — from breaking web apps to keeping enterprise data platforms up. Select a span to inspect it.

trace career · 4 spans · 6y

Jan 2025 — Presentcurrent

Customer Reliability Engineer (Infra)

Astronomer

On the call when enterprise Airflow platforms break — and the one who makes sure it doesn't happen the same way twice.

customer deployments
500+
customer deployments
less alert fatigue
25%
less alert fatigue
faster RCA & fix validation
20%
faster RCA & fix validation
identity compliance
100%
identity compliance
  • Lead incident calls with enterprise customers — deep dives into cloud infrastructure, networking and Airflow across AWS, GCP and Azure.
  • Design network architectures for 50+ customer environments: DNS, VPC peering, VPN, transit gateways and PrivateLink.
  • Own the Terraform repository for Chronosphere — hundreds of monitors, dashboards and notification policies, with manual drift eliminated.
  • Build Terraform + Helm setups that replicate customer deployments, cutting RCA and fix validation time by 20%.
  • Automate multi-cloud identity — Workload Identity Federation, IAM roles, service accounts — at 100% compliance.
  • Ship upstream fixes to Apache Airflow and OpenTelemetry while maintaining internal patches for zero-day mitigation.
  • Airflow
  • Terraform
  • Helm
  • Chronosphere
  • OpenTelemetry
  • AWS
  • GCP
  • Azure

02The startup · 2021 – 2023

Before the pager, I built a company.

Uprevol — my college startup: a jobs platform for India's daily-wage workforce.

Uprevol started in college with one question: why do daily-wage workers still find jobs by word of mouth? It began as a scrappy backend I wrote myself and grew into a real company — Flutter apps for workers and recruiters, a web app, a launch-ready API, a trust-and-safety console, and a team of 50+.

I wasn't only the engineer. I hired the team and trained it in batches, ran the drills, took the product to market, worked out how it would earn, and tested every release before it reached a user. Most of what I know about ownership, I learned here.

registered users
1,000+
registered users
people hired & trained
50+
people hired & trained
repositories in the org
13
repositories in the org

6 hats · 5 levels · 12 contributors · how the platform worked

03Projects

Things I built, drawn from the source.

Each one reads two ways: a quick overview anyone can follow, and a deep dive — with architecture modeled from the actual source — for those who want the details.

P1Terraform · AWS · Kubernetes · 2025

EKS Observability, end to end

Run one command and get a Kubernetes cluster on AWS that already watches itself — metrics, dashboards and alerts included.

command to stand it all up
1
command to stand it all up
alert rules out of the box
5
alert rules out of the box
manual clicks in Grafana
0
manual clicks in Grafana

How it works

  1. 1

    Build the cluster

    Terraform creates the network, an EKS cluster and low-cost spot worker nodes.

  2. 2

    Install the watchers

    Helm deploys Prometheus to collect metrics and Grafana to show them.

  3. 3

    Wire the alarms

    Rules catch crashing pods, restart loops and CPU spikes — and email the on-call.

  4. 4

    Draw the dashboards

    As soon as Grafana answers, Terraform creates the dashboards. Nobody clicks anything.

Open source

Fixes I sent upstream after hitting the bugs in production — OpenTelemetry metrics crashing Airflow's scheduler and DAG processor.

Smaller builds

all repositories on GitHub

04Security research

I used to break things for a living.

Before reliability, I spent two years in bug bounties. It taught me to read systems the way an attacker does — which is still how I review infrastructure.

vulnerabilities reported
25+
vulnerabilities reported
rank on Enjin's program
#8
rank on Enjin's program
HackerOne badges
3
HackerOne badges

HackerOne badges

  • Publish or PerishJun 2021
  • TrailblazerFeb 2021
  • InsecticideFeb 2021

Thanked by

  • Enjin2 / 2 validrank #8
  • JNJ Mobile1 reportrank #19
hackerone.com/er_salil
Publicly disclosedSeverity: LowResolved

Enjin · report #1091403

CSRF bypassed on logout endpoint

Cross-Site Request Forgery (CSRF)

  1. 01 · Find

    Enjin's logout endpoint accepted cross-site POST requests with no CSRF protection.

  2. 02 · Prove

    Any third-party page could silently sign a logged-in user out — shown with a self-submitting form.

  3. 03 · Fix

    Enjin's team confirmed it, shipped the fix and published the report.

<!-- shape of the PoC -->
<form action="https://target.example/logout" method="POST"></form>
<script>document.forms[0].submit()</script>
19 upvotesreported Jan 2021 · disclosed Jun 2022Read the report

05How I work

Opinions, earned on call.

  1. 1

    Alerts are code.

    If a monitor isn't in version control, it drifts. Hundreds of Chronosphere monitors live in Terraform for exactly that reason.

  2. 2

    Reproduce before you resolve.

    I rebuild the customer's setup with Terraform and Helm before touching theirs. It's how RCA got 20% faster.

  3. 3

    Fix it where it lives.

    A workaround helps one customer. An upstream patch helps everyone running Airflow.

  4. 4

    Page humans sparingly.

    Every alert costs attention. Cutting alert fatigue by 25% made the remaining pages worth answering.

Toolbox

Cloud & platform
  • AWS
  • GCP
  • Azure
  • Kubernetes
  • Docker
  • Terraform
  • Helm
  • ArgoCD
  • Ansible
  • CloudFormation
  • Jenkins
Observability
  • OpenTelemetry
  • Prometheus
  • Grafana
  • Chronosphere
  • ELK
  • New Relic
  • Dynatrace
  • Robusta
  • Graylog
  • Splunk
  • Instana
  • Sentry
Code & data
  • Python
  • FastAPI
  • React
  • SQL
  • PostgreSQL
  • MySQL
  • MongoDB
  • Couchbase
  • C / C++
Security
  • Vulnerability assessment
  • Security testing
  • Burp Suite

Certifications

  • Kong Gateway FoundationsKong2024
  • Full Stack Observability PractitionerNew Relic University2024
  • Linux AdministrationRed Hat2021
  • Network AssessmentPalo Alto Networks2021
  • Cybersecurity & Infrastructure SecurityU.S. DHS2021

Education

B.Tech, Computer Science

Anand International College of Engineering

2019 – 2023 · GPA 9.2 / 10

06Contact

Pager won't stop? Let's talk.

SRE, platform and observability roles, or a second pair of eyes on an incident you can't shake. The form goes straight to my inbox — it's a real API, so curl works too.

[email protected]Book a session on Topmate
POST /api/contact