If it is done twice, it becomes code
Terraform and Ansible over runbooks that depend on someone remembering the steps.
SRE · CLOUD · PLATFORM ENGINEERING
Building systems that are reliable by design.
I’m Ismael Vicencio, an SRE and platform engineer in Santiago, Chile. I started out writing backend software for insurance and toll roads, moved into platform work at Mercado Libre, and have spent the years since keeping Kubernetes platforms observable, affordable, and boring to operate.
ismael@ismaelvicencio.cl:~$ git log -1 --oneline
ef3523d Make the site specific: real work, build data…
ismael@ismaelvicencio.cl:~$ cat deploy.txt
- built
- 2026-10-13 01:48 UTC
- by
- github-actions
- runs on
- Cloudflare Workers
- client JS
- one 50-line file
ismael@ismaelvicencio.cl:~$
Chile gets earthquakes. Production gets incidents. The job is the same either way: design for the shake, measure it, and recover fast.
Selected work
Three systems I helped keep standing
Mall Plaza2024 – 2026
Making Kubernetes cheaper and calmer
Right-sized workloads pod by pod, then tuned autoscaling with HPA, VPA, and queue-depth metrics. Moved Java services to GraalVM and distroless images, and unified logs with Fluent Bit and Loki so an alert points at a cause instead of a symptom.
Mercado Libre2022 – 2024
Backend services at marketplace scale
Built and operated services in Java, Go, and Python on Google Cloud and AWS, handling millions of requests a day against availability SLOs. Gave product and infrastructure teams shared dashboards over BigQuery and Looker.
Kapsch TrafficCom2021 – 2022
Tolling that cannot miss a car
Led the backend for free-flow tolling in Chile and Ecuador. There are no barriers and nobody stops, so every passage has to be captured and processed in real time, whatever else is failing.
Working principles
Rules I keep coming back to
An SLO before a dashboard
Decide what “working” means for the user first. Alerts and graphs follow from that, not the other way around.
Right-size, then autoscale
Autoscaling a badly sized workload only scales the waste. Measure requests and limits before adding replicas.
No credentials in code
Secrets live in a vault, and the pipeline fails when one shows up in a repository.
Engineering notes
Things I’m learning and building
Technical write-ups, lab notes, troubleshooting guides, and lessons worth sharing.
How this site is built
A static Astro site on Cloudflare Workers: one Worker that almost never runs, a strict CSP, and a pipeline that deploys on every push.
About
From backend developer to SRE
I work in the space between software and infrastructure: how systems are provisioned, how changes reach production, and how teams understand what happens after deployment. I started as a software engineer and moved into infrastructure and platform work at Mercado Libre, then into SRE.
My experience covers multi-cloud platforms, Kubernetes, DevSecOps, internal developer platforms, CI/CD pipelines, and SLOs for critical services. I hold the Google Cloud Professional Cloud Architect and Associate Cloud Engineer certifications, plus Azure Administrator Associate (AZ-104) and Azure Fundamentals (AZ-900).
Contact
Let’s connect.
For engineering conversations, collaboration, or sharing ideas about cloud infrastructure and reliability.