Serdarcan Büyükdereli

Senior DevOps & Cloud Engineer

I keep production running on AWS and Kubernetes.

I have been running production infrastructure for more than nine years. These days I look after 20 customer accounts at LimonCloud, most of them on AWS. The work is migrations, cost reviews and finding the real cause when something breaks.

Portrait of Serdarcan Büyükdereli

About

I am an Electronics and Communication Engineer. I started with bare-metal servers, Proxmox, VMware and KVM virtualization and networking at a telecom operator. After that I worked on Kubernetes, CI/CD, GitOps and monitoring at BİTES, Insider, arabam.com and ICRYPEX, across AWS, Google Cloud and Huawei Cloud.

Today I work at LimonCloud, a cloud consultancy and managed services provider. My days are production support for 20 active customer accounts: incident root-cause work, EKS upgrades, an Azure AKS to EKS migration, cost and security reviews, and backup standards.

I also build internal tools in Python. Two of them are a read-only AWS inventory with 85 scanners and an AWS architecture and pricing web app with its own MCP server.

Work by area

The same work as in the timeline below, grouped by what it was for. Company names show where each item happened.

Migrations and upgrades

Moving production between platforms, accounts and versions.

  • Worked on a production migration from Azure AKS to EKS. After cutover I audited 73 deployments and 73 ingresses, fixed pods stuck in Pending, and resolved ALB ingress 404 and 500 errors.

    LimonCloud

  • Upgraded a 7-node EKS cluster from 1.33 to 1.36 in one night, with EBS snapshots taken first.

    LimonCloud

  • Led the split of a customer's workloads across two AWS accounts: 17 EC2 instances, 11 ECS services, RDS and 13 CodePipelines.

    LimonCloud

  • Led the migration from Elastic Beanstalk to ECS, with the new infrastructure defined in Terraform and Terragrunt.

    Insider

  • Moved Netsantral from a monolith to 8 microservices on Kubernetes. 13 unstable servers became 6 stable ones.

    Netgsm

  • Tested a move from x86 to AWS Graviton (ARM) for a customer's MongoDB replica set and ClickHouse cluster under load, with 3 x86 and 3 ARM nodes and a site-to-site VPN at 53 ms latency. MongoDB finished with zero errors, and the longest write stall during a planned primary change was 1.34 seconds. ClickHouse lost no data, with 41 insert errors in total while Keeper was being moved. This was a lab test, not a production move. It was delivered with a results report and a cutover and rollback draft.

    LimonCloud

  • Moved a customer's PHP application to AWS. The environment was built in Terraform: VPC, a single NAT gateway, ECS Fargate on ARM64, RDS MariaDB, ElastiCache, CloudFront with WAF, CloudTrail and KMS. Smoke tests passed after cutover, and the first month's usage cost came in about 11% below the calculator estimate.

    LimonCloud

  • Built the AWS environment in Terraform for an application that ran on one VMware VM (nginx, PostgreSQL 13, Redis, RabbitMQ and Go binaries). It includes CloudTrail, a budget alarm, 9 CloudWatch alarms, a tag-based AWS Backup plan, Pritunl VPN and Secrets Manager. The plan targets PostgreSQL 17 to avoid the PostgreSQL 13 Extended Support charge.

    LimonCloud

  • Preparing the move of an on-prem Docker Compose system with 118 containers to EKS: container inventory, 34 deployment manifests, 535 of 571 API gateway routes converted to Kubernetes service names, and a Terraform plan that adds 165 resources. In progress; the workloads have not been moved yet.

    LimonCloud

  • Side work for the AKS to EKS move: scanned 367 GitLab projects, inventoried 255 webhooks and updated the Jenkins target of 243 of them in bulk. An nginx server with 57 subdomains was moved from another cloud to AWS as an AMI.

    LimonCloud

Cost optimization

Measuring real usage first, then cutting what is not used.

  • CPU requests on a production EKS cluster were about 30x the p95 usage over 7 days of Prometheus data. The node group went from 19 nodes to 5, about 63% lower at list price.

    LimonCloud

  • Matched 22 on-demand EC2 instances to 1-year Reserved Instances. EC2 compute cost dropped by about 50% at list price.

    LimonCloud

  • Cleaned up two AWS accounts over two rounds. The first round alone cut the bill by 13%, and I proposed further savings more than twice the size of what was done.

    LimonCloud

  • Cut Google Cloud and Elasticsearch spend by 20% through log management and instance right-sizing.

    ICRYPEX

  • Built a Lambda function that checks tags across all resources and reports untagged ones to Slack for cost allocation.

    Insider

  • Moved an ElastiCache Serverless cache from Redis OSS to Valkey without downtime. The expected monthly cost drops by about 92%. On the same account, the CloudFront Pro plan was cancelled in favor of pay-as-you-go.

    LimonCloud

Incidents and root cause

Finding the actual cause, fixing it, and writing it down.

  • A 20-minute VPN outage. A DNS server reboot had dropped a non-persisted route, and an over-broad Pritunl NAT rule had hidden it for 18 days. Fixed the same morning.

    LimonCloud

  • Cloudflare 520 errors. A 75-second query was hitting a 60-second proxy timeout. Fixed on both HAProxy nodes with no downtime.

    LimonCloud

  • CloudFront 502 errors. An origin request policy dropped the Host header in front of a host-routed ALB.

    LimonCloud

  • Cut a compromised web server over to a clean rebuild: moved 19 ALB rules and removed 3 unauthorized admin accounts from the production database. TTFB went from 9.7 s to 1.5 s.

    LimonCloud

  • Changes, migrations and outages are written up as runbooks and incident reports.

    LimonCloud

Monitoring and observability

Metrics, logs and alerts that people act on.

  • Moved a production kube-prometheus-stack from emptyDir to persistent volumes and put Grafana behind ALB with ACM. The setup became the company's Helm standard.

    LimonCloud

  • Built the monitoring and alerting infrastructure on Google Cloud for GKE and other services.

    ICRYPEX

  • Designed a highly available monitoring system from scratch with Prometheus and Thanos, integrated with Consul, with alerts routed through Zenduty.

    arabam.com

  • Installed Zabbix on EC2 with alarms forwarded to Discord, and Prometheus with a custom HPA on DigitalOcean Kubernetes.

    BİTES

  • Wrote custom Prometheus node exporters in Python, created Alertmanager alerts, and integrated the ELK stack with Kubernetes.

    Netgsm

  • Reviewed CloudWatch alarms that were permanently red. Most were target tracking scale-in alarms, which is normal behavior. A memory scaling alarm that worked in reverse was fixed by rewriting its policy, and the total went from 26 alarms to 20.

    LimonCloud

Backup and recovery

A backup counts once the restore has been timed.

  • A customer's AWS Backup and DLM setup looked healthy but had taken zero backups. I rebuilt it as a tag-based, four-tier design. It became the company standard and now runs at two customers.

    LimonCloud

  • Restore rehearsal on that design took 1 min 17 s for EC2 and 8 min 28 s for RDS.

    LimonCloud

  • Tested Kasten K10 exports to Veeam on a 6-VM kubeadm cluster with Longhorn. Restores took 131 s from snapshot and 144 s from Veeam. Test environment, not production.

    LimonCloud

  • Rewrote S3 backup scripts from 2021 that were filling the SQL Server backup disk. The zip step was removed, and every step now checks its exit code and the object size in S3. 1,672 GB of backups were moved to S3 with verification, and the nightly backup now finishes in 8 to 11 minutes.

    LimonCloud

Security and reviews

Read-only reviews, secrets handling and access.

  • Scanned 781 resources across 18 regions for a customer, found internet-exposed domain controllers, and caught API Gateway drift in a 157-file Terragrunt repository that would have opened the production API to the internet on the next apply.

    LimonCloud

  • Wired IdentityServer SSO for an internal HR application through a separate ingress host so the live SSO host stayed untouched, and flagged plaintext secrets in pipeline variable groups to be rotated.

    LimonCloud

  • Deployed a highly available HashiCorp Vault cluster on-premise as the central secrets store.

    arabam.com

  • Automated base image builds pushed to Amazon ECR, with vulnerability scan results delivered over Slack.

    Insider

Platform, CI/CD and GitOps

Clusters, pipelines and infrastructure as code.

  • Built a new EKS 1.36 cluster in Terraform with a private endpoint, arm64 nodes, AWS Load Balancer Controller and Cluster Autoscaler.

    LimonCloud

  • Run Terraform and Terragrunt with per-tenant state and mandatory cost-allocation tagging.

    LimonCloud

  • Created a declarative GitOps workflow with Azure DevOps (CI) and ArgoCD (CD) on GKE, then extended it to Huawei Cloud CCE so both clouds shared one delivery model.

    ICRYPEX

  • Created standard Helm templates following 12-Factor App principles and automated releases with GitLab CI and ArgoCD.

    arabam.com

  • Wrote a Helm chart template for automatic on-prem setups, saving 40% admin time, and used Ansible for installs on air-gapped networks.

    BİTES

  • Set up CodeBuild-based CI/CD for an ECS environment: a pipeline with a manual approval step before deploy, 6 CodeBuild projects and rollback. It was later split into separate test and production pipelines, and ECS autoscaling was added with step scaling.

    LimonCloud

  • Renewed the certificates of an RKE2 cluster node by node, draining each one at HAProxy first. Moved 17 Azure DevOps service connections to namespace-scoped service account tokens.

    LimonCloud

  • Installed Apache Superset and an MCP server on EKS, with Valkey, cert-manager and ALB ingresses. Every container has resource limits and secrets come from 1Password.

    LimonCloud

  • Made applications see the real client IP of users who come in through the VPN: PROXY protocol on HAProxy and the ingress, plus a persistent NAT exemption on the VPN server. Tested end to end on 31 ingress hosts.

    LimonCloud

Databases

Moves, repairs and performance work on data stores.

  • Moved two single-node Elasticsearch clusters to two nodes each and a MongoDB replica set to a new server. The move was a cold copy in a customer-approved evening window, and MongoDB was verified by comparing counts across 7,385 collections. Prometheus exporters and Grafana dashboards were added afterwards.

    LimonCloud

  • Repaired a replica that was stuck on an old timeline in a 3-node Patroni PostgreSQL cluster, with no downtime: enabled pg_rewind and reinitialized two replicas. Both ended up streaming with zero lag, and the 13 databases matched on all three nodes.

    LimonCloud

  • Traced an IO bottleneck on two PostgreSQL servers to the IOPS and throughput limits of a single gp3 volume. The volumes were raised to 7,000 IOPS and 350 MB/s without downtime, and peak usage on one server dropped to 57%.

    LimonCloud

AI and LLM work

MCP servers, agent skills and retrieval, used in day-to-day operations.

  • Wrote the MCP server behind AWS Architecture Studio. It talks to 14 AI providers over plain HTTP and has a local-model option. Its tools carry read-only and open-world hints, and remote mode refuses local file paths.

    LimonCloud

  • Wrote Claude Code skills that the team uses for customer discovery, cost analysis and reporting.

    LimonCloud

  • Designed two target AWS architectures for a prospect, with Amazon Bedrock's EU inference profile to keep data in the EU. Proposal stage, not implemented.

    LimonCloud

  • Ran an MCP server on EKS next to Apache Superset. Every container has resource limits and secrets come from 1Password.

    LimonCloud

  • local-rag: a retrieval pipeline over my own notes that runs fully on my machine, with Qdrant, hybrid retrieval, reranking and a local Gemma model through Ollama. I wrote an evaluation set by hand and compared embedding models on Turkish queries.

    Side project

Internal tools

Tools I wrote because the existing ones left a gap.

  • AWS Architecture Studio: a web app and MCP server that turns an architecture description into a shareable calculator.aws estimate and a draw.io diagram.

    LimonCloud

  • A read-only AWS inventory script in Python (Boto3) with 85 scanners. Its output feeds architecture, security and cost reviews.

    LimonCloud

  • Jira integrations: a ticket sync daemon, a comment sender that refuses to post until the text passes a terminology check, and an exporter that documented 123 automation rules.

    LimonCloud

  • A customer-memory CLI over SQLite FTS5. Credentials are injected from 1Password at run time, and a pre-write hook blocks secrets from reaching documents.

    LimonCloud

  • A read-only discovery tool that runs every 14 days against AWS, Kubernetes, vCloud and servers reached over SSH. It reports what changed since the last run, cost, CVEs and security findings, and produces a masked PDF for the customer. Five customers were scanned in November 2026.

    LimonCloud

Experience

May 2026 - Present

Ankara

LimonCloud

Senior DevOps & Cloud Engineer

Cloud consultancy and managed services provider. 20 active customer accounts, mostly AWS, with some Google Cloud, vCloud Director and on-prem Kubernetes.

  • Measured CPU requests on a production EKS cluster against 7 days of Prometheus data and found them at about 30x p95 usage. The node group was then consolidated from 19 to 5 nodes, about 63% lower at list price.
  • Matched 22 on-demand EC2 instances to 1-year Reserved Instances for the same customer, cutting EC2 compute cost by about 50% at list price.
  • Worked on a production migration from Azure AKS to EKS: audited 73 deployments and 73 ingresses after cutover, fixed pods stuck in Pending by removing stale nodeAffinity rules, resolved ALB ingress 404 and 500 errors from subnet and path-priority conflicts, rebuilt a missing service with deployment, service, ingress and HPA manifests, and moved Jaeger to EKS as a Helm release.
Show all 12 itemsShow fewer
  • Cleaned up two AWS accounts for another customer over two rounds. The first round alone cut the bill by 13%, and I proposed further savings more than twice the size of what was done.
  • Upgraded a 7-node EKS cluster from 1.33 to 1.36 in one night, taking EBS snapshots first and fixing a lost IRSA annotation on the EBS CSI driver. Built a new EKS 1.36 cluster in Terraform with a private endpoint, arm64 nodes, AWS Load Balancer Controller and Cluster Autoscaler.
  • Moved a production kube-prometheus-stack from emptyDir to persistent volumes and put Grafana behind ALB with ACM. The setup became the company's Helm standard.
  • Root-caused production incidents: a 20-minute VPN outage after a DNS server reboot dropped a non-persisted route, hidden for 18 days by an over-broad Pritunl NAT rule and fixed the same morning; Cloudflare 520s from a 75-second query hitting a 60-second proxy timeout, fixed on both HAProxy nodes with no downtime; CloudFront 502s from an origin request policy that dropped the Host header in front of a host-routed ALB.
  • Cut a compromised web server over to a clean rebuild: moved 19 ALB rules, removed 3 unauthorized admin accounts from the production database, and TTFB went from 9.7 s to 1.5 s.
  • Led the split of a customer's workloads across two AWS accounts (17 EC2 instances, 11 ECS services, RDS, 13 CodePipelines), and changed the timezone on 4 SQL and application servers through SSM with about 40 seconds of downtime.
  • Took an internal HR application (Next.js web, .NET API) through its production deployment on Kubernetes with Azure DevOps release pipelines, wired its IdentityServer SSO (OIDC client with access and role scopes) through a separate ingress host so the live SSO host stayed untouched, and flagged plaintext secrets in the pipeline variable groups to be marked secret and rotated.
  • Ran Terraform and Terragrunt with per-tenant state and mandatory cost-allocation tagging.
  • Mentor junior and mid-level engineers with day-to-day guidance and task support, and write changes, migrations and outages up as runbooks and incident reports.

April 2025 - January 2026

Ankara

ICRYPEX

Senior DevOps Engineer

  • Built the monitoring and alerting infrastructure on Google Cloud for GKE and other services.
  • Created a declarative GitOps workflow with Azure DevOps (CI) and ArgoCD (CD) on Google Kubernetes Engine, then extended it to Huawei Cloud CCE so both clouds shared one delivery model.
  • Cut Google Cloud and Elasticsearch spend by 20% through log management and instance right-sizing.
  • Worked across Huawei Cloud (CCE, ECS) and Google Cloud (GKE, Cloud Run, Logging, Monitoring), and integrated and managed RabbitMQ, PostgreSQL, Kafka and Redis.

September 2023 - April 2025

Ankara

arabam.com

Senior DevOps Engineer

  • Designed a highly available monitoring system from scratch with Prometheus and Thanos, integrated Prometheus with Consul, and routed alerts through Zenduty to improve incident response.
  • Created standardized Helm templates following 12-Factor App principles and automated releases with full GitOps delivery on GitLab CI and ArgoCD.
  • Deployed a high-availability HashiCorp Vault cluster on-premise as the central secrets store and aligned application secrets handling with 12-Factor App principles.
Show all 6 itemsShow fewer
  • Managed a Ceph cluster providing S3 object storage, CephFS and RBD, ran PostgreSQL, Kafka and Redis, and ran Airflow for scheduled jobs and reports with HashiCorp Vault integration.
  • Worked with Docker, Kubernetes, Helm, Ansible, Terraform and Keycloak, and managed AWS infrastructure with Terraform and Terragrunt.
  • Evaluated a service mesh in the non-production environment and mentored junior engineers with guidance and task support.

December 2021 - September 2023

Ankara

Insider

DevOps Engineer

  • Led the migration from Elastic Beanstalk to AWS ECS. The new ECS infrastructure was defined in Terraform and Terragrunt.
  • Wrote Dockerfiles and build scripts for PHP Laravel and Go applications and deployed them to ECS with CodePipeline and CodeBuild.
  • Managed EKS with Helm and built an EKS test cluster with Terragrunt.
Show all 7 itemsShow fewer
  • Built a Lambda function that checked tags across all resources and reported untagged resources to Slack for cost allocation.
  • Automated base image builds pushed to Amazon ECR, with ECR vulnerability scan results delivered over Slack.
  • Managed AWS (EC2, ECS, S3, ACM, Elastic Beanstalk, Lambda, EKS, Route53, CloudWatch).
  • Datadog monitoring.

October 2020 - December 2021

Ankara

BİTES - Defense & Aerospace Technologies

DevOps Engineer

  • Dockerized Xpermeet and ran it on DigitalOcean Kubernetes, with Prometheus monitoring and a custom HPA.
  • Wrote a Helm chart template for automatic on-prem setups, saving 40% admin time.
  • Used Ansible for installations in customer environments on restricted, air-gapped networks.
  • Installed a Zabbix monitoring system on EC2, with Zabbix alarms forwarded to Discord.
  • Worked with Docker, Kubernetes, Helm, Envoy, PostgreSQL, MinIO and Redis, and managed AWS (EC2, ECS, S3, ACM, CloudFront, Route53, CloudWatch).

January 2018 - September 2020

Ankara

Netgsm Communication and Information Technologies Inc.

DevOps Engineer

  • Moved Netsantral from a monolithic architecture to Docker and Kubernetes as 8 microservices with Macvlan networking, reducing 13 unstable servers to 6 stable servers and adapting it to on-premise installations through Kubernetes.
  • Integrated PHP and Java projects with GitLab CI/CD pipelines running on Kubernetes.
  • Monitored Kubernetes cluster health with Prometheus and Grafana, wrote custom Prometheus node exporters in Python, and created Alertmanager alerts for developers and the DevOps team.
  • Integrated the ELK stack with Kubernetes and routed high-volume UDP logs through Redis and Rsyslog to reduce processing complexity.
  • Managed Proxmox, VMware and KVM virtualization, routers, switches, firewalls and mail servers; Bash and Python scripting.

October 2017 - December 2017

Ankara

Netgsm Communication and Information Technologies Inc.

Intern

  • Network and Linux.

Projects

Projects from my current work and my own time. Customer names are left out.

LimonCloud

AWS Architecture Studio

Internal web app and MCP server that turns an architecture description into a shareable calculator.aws estimate and a draw.io diagram. I built it because the vendor AWS MCP servers the team uses (documentation, pricing) do not cover that gap.

Python, Docker and Traefik, with image versioning and smoke tests in Bitbucket Pipelines. It talks to 14 AI providers over plain HTTP and has a local-model option. MCP tools carry read-only and open-world hints, and remote mode refuses local file paths.

LimonCloud

Jira Integrations

A daemon that syncs assigned tickets into an Obsidian knowledge base, a comment sender that refuses to post until the text passes a terminology check, and an exporter that pulls Jira Cloud automation rules through the automation gateway API because the REST API does not expose them (123 rules documented).

LimonCloud

Security and Infrastructure Review

For a customer: scanned 781 resources across 18 regions, found internet-exposed domain controllers, and caught API Gateway drift in a 157-file Terragrunt repository that would have opened the production API to the internet on the next apply.

LimonCloud

Internal Application Inventory

Mapped 11 custom applications for a customer: stack, databases, Redis, RabbitMQ, object storage, SSO and integration dependencies.

LimonCloud

Tag-Based AWS Backup Reference Architecture

A customer's AWS Backup and DLM setup looked healthy but had taken zero backups. I rebuilt it as a tag-based, four-tier design that became the company standard and is now deployed at two customers. Restore rehearsal took 1 min 17 s for EC2 and 8 min 28 s for RDS.

LimonCloud

AWS Inventory and Discovery Tooling

Read-only AWS inventory script in Python (Boto3) with 85 scanners that maps resources across accounts without touching them. The output feeds architecture, security and cost reviews.

LimonCloud

Kubernetes Backup Proof of Concept

Built a 6-VM kubeadm cluster with Longhorn and CloudNativePG, MongoDB and MySQL operators to test Kasten K10 exports to Veeam. Restores took 131 s from snapshot and 144 s from Veeam. Test environment, not production.

LimonCloud

Internal Tooling and Secret Handling

Customer-memory CLI over SQLite FTS5 and Claude Code skills for discovery, cost analysis and reporting. Credentials are injected from 1Password at run time, and a pre-write hook blocks secrets from reaching documents.

LimonCloud

Target Architecture with Amazon Bedrock

Designed two target AWS architectures for a prospect, with Amazon Bedrock's EU inference profile to keep data in the EU. Proposal stage, not implemented.

LimonCloud

Graviton Migration Lab Test

Tested a move from x86 to AWS Graviton (ARM) for a customer's MongoDB replica set and ClickHouse cluster under load, with 3 x86 and 3 ARM nodes and a site-to-site VPN at 53 ms latency.

MongoDB finished with zero errors, and the longest write stall during a planned primary change was 1.34 seconds. ClickHouse lost no data, with 41 insert errors in total while Keeper was being moved.

This was a lab test, not a production move. It was delivered with a results report and a cutover and rollback draft.

LimonCloud

PHP Application Migration to AWS

Moved a customer's PHP application to AWS. The environment was built in Terraform: VPC, a single NAT gateway, ECS Fargate on ARM64, RDS MariaDB, ElastiCache, CloudFront with WAF, CloudTrail and KMS. Smoke tests passed after cutover, and the first month's usage cost came in about 11% below the calculator estimate.

LimonCloud

AWS Environment for a Single-VM Application

Built the AWS environment in Terraform for an application that ran on one VMware VM (nginx, PostgreSQL 13, Redis, RabbitMQ and Go binaries). It includes CloudTrail, a budget alarm, 9 CloudWatch alarms, a tag-based AWS Backup plan, Pritunl VPN and Secrets Manager. The plan targets PostgreSQL 17 to avoid the PostgreSQL 13 Extended Support charge.

LimonCloud

Docker Compose to EKS Migration Prep

Preparing the move of an on-prem Docker Compose system with 118 containers to EKS: container inventory, 34 deployment manifests, 535 of 571 API gateway routes converted to Kubernetes service names, and a Terraform plan that adds 165 resources. In progress; the workloads have not been moved yet.

LimonCloud

AKS to EKS Migration, Supporting Work

Side work for the AKS to EKS move: scanned 367 GitLab projects, inventoried 255 webhooks and updated the Jenkins target of 243 of them in bulk. An nginx server with 57 subdomains was moved from another cloud to AWS as an AMI.

LimonCloud

CodeBuild Pipelines for ECS

Set up CodeBuild-based CI/CD for an ECS environment: a pipeline with a manual approval step before deploy, 6 CodeBuild projects and rollback. It was later split into separate test and production pipelines, and ECS autoscaling was added with step scaling.

LimonCloud

RKE2 Certificate Renewal

Renewed the certificates of an RKE2 cluster node by node, draining each one at HAProxy first. Moved 17 Azure DevOps service connections to namespace-scoped service account tokens.

LimonCloud

Apache Superset on EKS

Installed Apache Superset and an MCP server on EKS, with Valkey, cert-manager and ALB ingresses. Every container has resource limits and secrets come from 1Password.

LimonCloud

Real Client IP Behind VPN

Made applications see the real client IP of users who come in through the VPN: PROXY protocol on HAProxy and the ingress, plus a persistent NAT exemption on the VPN server. Tested end to end on 31 ingress hosts.

LimonCloud

Elasticsearch and MongoDB Relocation

Moved two single-node Elasticsearch clusters to two nodes each and a MongoDB replica set to a new server. The move was a cold copy in a customer-approved evening window, and MongoDB was verified by comparing counts across 7,385 collections. Prometheus exporters and Grafana dashboards were added afterwards.

LimonCloud

Patroni PostgreSQL Replica Repair

Repaired a replica that was stuck on an old timeline in a 3-node Patroni PostgreSQL cluster, with no downtime: enabled pg_rewind and reinitialized two replicas. Both ended up streaming with zero lag, and the 13 databases matched on all three nodes.

LimonCloud

PostgreSQL Disk IO Bottleneck

Traced an IO bottleneck on two PostgreSQL servers to the IOPS and throughput limits of a single gp3 volume. The volumes were raised to 7,000 IOPS and 350 MB/s without downtime, and peak usage on one server dropped to 57%.

LimonCloud

SQL Server Backup Scripts

Rewrote S3 backup scripts from 2021 that were filling the SQL Server backup disk. The zip step was removed, and every step now checks its exit code and the object size in S3. 1,672 GB of backups were moved to S3 with verification, and the nightly backup now finishes in 8 to 11 minutes.

LimonCloud

ElastiCache Move to Valkey

Moved an ElastiCache Serverless cache from Redis OSS to Valkey without downtime. The expected monthly cost drops by about 92%. On the same account, the CloudFront Pro plan was cancelled in favor of pay-as-you-go.

LimonCloud

CloudWatch Alarm Cleanup

Reviewed CloudWatch alarms that were permanently red. Most were target tracking scale-in alarms, which is normal behavior. A memory scaling alarm that worked in reverse was fixed by rewriting its policy, and the total went from 26 alarms to 20.

LimonCloud

Periodic Discovery Tool

A read-only discovery tool that runs every 14 days against AWS, Kubernetes, vCloud and servers reached over SSH. It reports what changed since the last run, cost, CVEs and security findings, and produces a masked PDF for the customer. Five customers were scanned in November 2026.

ICRYPEX

GitOps for GKE and Huawei Cloud

Created a declarative GitOps workflow with Azure DevOps (CI) and ArgoCD (CD) on Google Kubernetes Engine, then extended the same model to Huawei Cloud CCE so both clouds shared one delivery path.

ICRYPEX

Google Cloud Monitoring Infrastructure

Built the monitoring and alerting infrastructure on Google Cloud to watch the health of GKE and other services.

ICRYPEX

Cloud Cost Optimization

Cut Google Cloud and Elasticsearch spend by 20% through log management and instance right-sizing.

arabam.com

Helm Templates and GitOps Pipeline

Created standardized Helm templates that follow 12-Factor App principles. Releases were automated with GitLab CI and ArgoCD.

arabam.com

Prometheus and Thanos Monitoring System

Designed a highly available monitoring system from scratch with Prometheus and Thanos, and integrated Zenduty for alert routing.

arabam.com

HashiCorp Vault Integration

Deployed a high-availability Vault cluster on-premise as the central secrets store and aligned application secrets handling with 12-Factor App principles.

Insider

PHP and Go Applications on AWS ECS

Wrote Dockerfiles and build scripts for PHP Laravel and Go applications.

The applications were deployed with CodePipeline and CodeBuild.

The ECS cluster was created with Terraform and Terragrunt.

Insider

Resource Tag Check for Cost Allocation

A Lambda function that checked the tags of all resources and sent the untagged ones to Slack.

Insider

Base Image Builder

Base images were built from Dockerfiles and pushed to Amazon ECR automatically. ECR vulnerability scan results were sent to Slack.

Insider

EKS Test Deployment

Created an EKS cluster with Terragrunt and deployed the application on it with Helm.

BİTES

Xpermeet on Kubernetes with Helm

Dockerized Xpermeet and ran it on DigitalOcean Kubernetes with Prometheus monitoring and a custom HPA. A Helm chart template for automatic on-prem setups saved 40% admin time.

BİTES

Zabbix Monitoring System

Installed Zabbix on an EC2 server and forwarded its alarms to Discord.

Netgsm

Netsantral on Docker and Kubernetes

Netsantral was moved from a monolithic architecture to containers and split into 8 microservices on Kubernetes.

Macvlan networking was integrated with Kubernetes.

13 unstable servers were reduced to 6 stable ones, and the product was adapted to on-premise installations.

Netgsm

GitLab CI/CD on Kubernetes

Integrated PHP and Java projects with GitLab CI/CD pipelines running on Kubernetes.

Netgsm

Prometheus Monitoring System

Monitored the health of Kubernetes clusters with Prometheus and Grafana, wrote custom node exporters in Python, and created Alertmanager alerts for developers and the DevOps team.

Netgsm

ELK Log System

Integrated the ELK stack with Kubernetes and moved high-volume UDP logs through Redis and Rsyslog to reduce processing complexity.

Side project

theinfinity.dev

Technical blog with its own publishing pipeline on the same host, deployed and verified through a scripted release flow.

Visit

Side project

ParaAlfa

Fund and deposit-rate return tracker: Next.js frontend, Python data pipeline, Playwright-based collection. It ran in production behind Traefik from July to November 2026.

I handled SEO, sitemap and structured data end to end, and monitored it through Search Console and GA4.

Two FastAPI services exposed the market data and a portfolio tracker as REST APIs with Swagger docs, backed by SQLite, with a Pytest suite and an MCP server, deployed with Docker Compose. The portfolio API uses Pydantic models and authenticates requests with bearer sessions or per-user API keys.

Side project

PetRadar

Price discovery and tracking tool built for a small retail business: FastAPI, Pydantic and Pytest. It keeps each site's full catalog locally from its sitemap and tracks prices daily.

Side project

local-rag

Fully local retrieval-augmented generation (RAG) pipeline over a personal knowledge base: Docling chunking, Qdrant, hybrid retrieval (dense embeddings plus BM25) fused with RRF, cross-encoder reranking, and generation on a local Gemma model through Ollama.

I built a hand-written evaluation set to measure retrieval quality, compared embedding models on Turkish queries, and exposed the search to Claude Code as an MCP server.

Open source and writing

open-source-tools

A public list of open-source projects and tools that I maintain on GitHub.

Visit

Open source and writing

techweekly

A weekly curated collection of tools, libraries and articles in software development, DevOps and AI, published by The Infinity.

Visit

Open source and writing

Medium

Articles I publish on Medium.

Visit

Skills

Cloud
AWS (EC2, ECS, EKS, Lambda, S3, RDS, CloudFront, Route53, CloudWatch, Backup, SSM, Cost Explorer), Azure (AKS, Azure DevOps), Google Cloud (GKE, Cloud Run, Monitoring), Huawei Cloud (CCE, ECS), DigitalOcean, vCloud Director
Containers and orchestration
Kubernetes, EKS, GKE, RKE2, kubeadm, Helm, Docker, ECS Fargate
Infrastructure as code and automation
Terraform, Terragrunt, Ansible, Bash, Python (Boto3)
CI/CD and GitOps
GitLab CI, Azure DevOps, ArgoCD, Flux, CodePipeline, CodeBuild, Jenkins, Bitbucket Pipelines, GitHub Actions (personal projects); builds and deployment of Go and PHP services (Dockerfiles, build scripts)
Observability
Prometheus, Thanos, Grafana, Loki, OpenTelemetry, Jaeger, SigNoz, ELK, Zabbix, CloudWatch, Datadog, Alertmanager, Zenduty
Backup and DR
AWS Backup, DLM, EBS snapshots; Kasten K10, Longhorn and Veeam (test environment)
Data, storage and messaging
SQL, PostgreSQL, Kafka, RabbitMQ, Redis, Elasticsearch, MinIO, Ceph, Airflow
Networking and load balancing
Routers, switches and firewalls (MTCNA, MTCIPv6E), Pritunl VPN, HAProxy, keepalived, Nginx, Traefik, Envoy, AWS ALB, Cloudflare
Security and secrets
HashiCorp Vault, Keycloak, 1Password CLI, IAM, WAF (design and findings)
Linux and virtualization
Linux, Proxmox, VMware, KVM
Backend (personal projects)
FastAPI, Pydantic, Pytest, SQLite
AI automation
Claude Code, Codex, MCP servers, agent skills, LLM-assisted operations tooling
Practices
Incident root-cause analysis, runbooks and incident reports, disaster recovery, capacity planning, DevSecOps, platform engineering, mentoring

Certifications and education

Certifications and courses

  • AWS Certified Generative AI Developer - Professional

    2026, valid until 2029Verify on Credly

  • AWS Certified Security - Specialty

    2026, valid until 2029Verify on Credly

  • Certified Kubernetes Administrator (CKA)

    2023, expired 2026

  • White Hat Hacker

    2019

  • MTCNA

    2018

  • MTCIPv6E

    2018

  • Work and Travel

    2017

Education

Süleyman Demirel University, Isparta

B.Sc. in Electronics and Communication Engineering

September 2012 - January 2018

Languages

  • Turkish, Native
  • English, Professional working proficiency

Interests

  • AI, LLM and AI agents
  • Building side projects with Codex, Claude and Antigravity

Get in touch

The quickest way is a calendar invite. Pick a time that suits you and add me as a guest; 30 minutes is usually enough for a first call. I am in Ankara (UTC+3).