Juan Tamariz

KEY CONTRIBUTIONS
• Set up the infrastructure for 4 different environments, including production. CI/CD, monitoring, backups, and security tools.
• Established the automation to continuously introduce security patches from lower environments to production.
• Set up AWS Inspector to validate possible new vulnerabilities in the code.
• Set up a CI/CD pipeline including code testing, security assessment, and a no-downtime deployment strategy that covers database upgrades. This reduced the application's downtime in production and increased the production release frequency.
• Implemented core component upgrades to reduce costs and maximize performance for the client.

I built a framework to automate the bootstrapping of several tools for a CI/CD pipeline, which included the build, code promotion, static code testing, performance baselines, and continuous deployment.

The impact of my work was a significant reduction of implementation time for new pipelines from 30 days to 7 days.

Using KubeSpawner, I developed an implementation of JupyterHub Notebooks. It provides a long-term solution for on-demand autoscaling of user instances alongside Kubernetes' allocated resources.

At a glance, for every user logged in, a new Kubernetes pod is created on-demand. When more resources are needed, the Kubernetes cluster will also auto-scale.

A full project that involved analyzing requirements and defining dependencies and stakeholders.

I created an Agile project to track the creation of every Terraform module. We ended up on a set of authorized scripts that were instanced on several projects on Google Cloud.

As a result, project setup speed improved from a month to every 3 days with the scripts. The framework considered the usage of the latest available Terraform version, along with a shared back end/state to make collaboration easier.

A Python platform library that wraps Pulumi with company-specific standards for networking, IAM, secrets, observability, and Datadog monitor automation. Engineers self-serve their own infrastructure by instantiating a single class; DevOps owns the library and the guardrails behind it.

OUTCOME
• 100% adoption across 14 microservices and 28 environments.
• Removed DevOps from the critical path of new service onboarding.
• IAM scoping, tagging, and Datadog checks are now applied uniformly across the estate with no human review needed for standard cases.
• Production-grade today; actively extended as new services come online.

Migrated 14 microservices across 28 environments from standalone Jenkins pipelines to a GitOps model on Argo CD, with PR-based promotion, automated sync, and full visibility into deployment health.

OUTCOME
• Deploy time dropped from 4 hours to 15 minutes.
• Deployment-related incidents reduced to zero (100% reduction).
• CI cost reduced by retiring legacy Jenkins workers.
• Every release is now auditable through Git history rather than tickets, and rollbacks are a single revert.

Designed and rolled out a Kafka cluster to replace an AWS DMS, Kinesis, and Glue data-movement pipeline. The new architecture consolidated 3 managed services into a single operator-friendly platform, improved end-to-end latency, and simplified the data team's day-to-day operations.

OUTCOME
• Around $7,000 per month in cloud cost savings.
• Lower operational surface for the data platform team.
• Foundation is now in place for downstream stream-processing work—including change data capture, real-time analytics, and event-driven services.

Migrated 3 EKS clusters with around 50 nodes from overprovisioned static node groups to Karpenter without any production disruption. I introduced PodDisruptionBudgets, topology spread constraints, and per-workload validation before cutover, then drove a 6-week rollout, culminating in a final FinOps validation report.

OUTCOME
• Around $4,000 per month in cluster cost savings, validated by the FinOps platform.
• EKS version upgrades are now non-disruptive—a recurring source of toil was eliminated.
• The same runbook was reused on a second client engagement, replacing the AWS Cluster Autoscaler and avoiding a Cast AI license.