Open to infra roles and contracts
Abdul Hadi Engineering infrastructure for Kubernetes|
I build cloud platforms that don't just run, they perform. Kubernetes, AWS, Azure, private clouds and the pipelines that ship to them.
-
AWS -
Azure -
Google Cloud -
OpenStack -
Kubernetes -
OpenShift -
Docker -
Helm -
Istio -
Argo CD -
Terraform -
Ansible -
Jenkins -
GitHub Actions -
Prometheus -
Grafana -
Elasticsearch -
Kibana -
Vault -
Trivy -
SonarQube -
Red Hat -
Cloudera -
Hadoop -
Spark -
Kafka -
Cassandra -
Ceph -
PostgreSQL -
Redis -
Linux -
Ubuntu -
NGINX -
Python -
Ollama
In numbers
- IOPS on EKS storage, up from 1,500
- 127k
- faster local LLM tokens on AWS CPUs
- 10.8×
- Kubernetes security cases in my research
- 18
- Certified Kubernetes Administrator
- CKA
01 / Selected work
Things I built, measured, and wrote down.
- Research 2026
Secure but Broken?
Do LLM-written fixes for Kubernetes security misconfigurations still work once deployed? Every repair is applied to a fresh cluster and has to pass seven gates: scanner rules, no new findings, server-side dry run, readiness, real user journeys, a load test against a measured baseline, and the original security intent checked by policy and a runtime probe.
Pre-registered protocol, 18 cases, full replication package with DOI.
- Kubernetes
- Python
- Terraform
- Amazon Bedrock
- EKS
- DynamoDB
repair #01 · fresh clusterrunning- 01 Scanner rules
- 02 No new findings
- 03 Server dry-run
- 04 Readiness
- 05 User journeys
- 06 Load vs baseline
- 07 Security intent
- Production 2025
EKS storage, 1,500 → 127,000 IOPS
Pods were slow while CPU and memory sat idle: the disk was the bottleneck. I ran a head-to-head on EKS between gp3 EBS on m5.large and local NVMe on a tainted i4i.xlarge node group, then wrote down the pitfalls so the next team skips them.
83× more IOPS, 95× lower latency.
- AWS EKS
- EBS CSI
- i4i NVMe
- Taints & tolerations
benchmark pod · same EKS clustertainted node group - AI infra 2025
Local LLMs on AWS CPU instances
Can a 7B model act as a senior sysadmin bot on a 4 vCPU m5.xlarge with no GPU? I benchmarked Mistral 7B and DeepSeek 7B on stock Ollama with forced cold starts, then rebuilt the runtime from source for the CPU it actually runs on.
2.5 → 27 tokens per second.
- EC2 m5.xlarge
- Ollama
- Mistral 7B
- Python benchmarks
m5.xlarge · 4 vCPU · no GPUsame promptstock Ollama2.5 t/s
built for this CPU27 t/s
prompt: “You are a Senior SysAdmin. A junior dev deleted /var/www/html.” · reply text illustrative, speeds measured
- Workshop 2026
Serverless AI study assistant
An agent on Amazon Bedrock (Nova Pro, Converse API tool loop) that stores courses in DynamoDB and manages a real Todoist list. Built as a hands-on workshop for first and second year CS students, with a least-privilege IAM policy and tests for every blank.
Runs end to end with one IAM user and one stack.
- Amazon Bedrock
- Lambda
- DynamoDB
- Streamlit
- Python
request pathConverse API tool loopleast-privilege IAM · one stack · pytest for every workshop blank
- Automation 2025
Kubernetes clusters from bare VMs
Role-based Ansible playbooks that turn fresh Ubuntu machines into a working kubeadm cluster, tested with Molecule against Docker, Vagrant and EC2.
Same cluster, three environments, one command.
- Ansible
- kubeadm
- Molecule
- Vagrant
- EC2
ansible-playbook site.ymlmolecule testdockervagrantec2 common ·ok·ok·ok master ·ok·ok·ok worker ·ok·ok·okPLAY RECAP · waiting
More from the field
Full list ↗Kubernetes & containers
-
Multi-node Kubernetes cluster
Fault-tolerant kubeadm cluster with RBAC, persistent storage and Prometheus/Grafana observability.
kubeadm · RBAC · Prometheus
-
OpenShift for enterprise apps
Red Hat OpenShift clusters for secure multi-tenant workloads, with autoscaling and strict security contexts.
OpenShift · SCC · Multi-tenant
-
Cluster bootstrapping with Ansible
Playbooks that provision networking, security and monitoring the same way on every node. Setup time down 80%.
Ansible · Terraform
-
Log analytics on Kubernetes
Elasticsearch, Fluentd and Kibana on Kubernetes, with Grafana dashboards for cluster health.
EFK · Grafana
-
Cassandra & Spark on Kubernetes
Cassandra StatefulSets and Spark workers with tuned resource limits and HDFS integration.
StatefulSets · Spark
Cloud, DevOps & security
-
DeepSeek & Mistral on AWS
Private LLM hosting on c5.2xlarge (CPU) and g4dn.xlarge (GPU), with Ollama compiled for the host CPU.
EC2 · g4dn · Ollama
-
Jenkins CI/CD with Docker
Automated test and release pipeline with Jenkins, Docker and GitHub, for consistent production builds.
Jenkins · Docker
-
Data security with IBM Guardium
Database activity monitoring across an enterprise fleet: real-time audit for GDPR and PCI DSS, and masking with no app changes.
DAM · GDPR · PCI DSS
Cloudera & big data
-
Cloudera CDP health check & tuning
Diagnosed HDFS and YARN bottlenecks on CDP clusters and tuned them for stability and faster queries.
CDP · HDFS · YARN
-
Cloudera upgrade on Azure Data Lake
Migrated and upgraded Cloudera workloads to Azure Data Lake under enterprise governance.
Azure · ADLS
-
IBM Storage Ceph
Multi-node Ceph for block and object storage, CRUSH maps tuned for throughput, 99.99% uptime.
Ceph · CRUSH · HA
02 / Toolchain
What I build with, day to day.
Cloud
- AWS
- Azure
- Google Cloud
- OpenStack
Containers
- Kubernetes
- OpenShift
- Docker
- Helm
- Istio
- Argo CD
IaC & delivery
- Terraform
- Ansible
- Jenkins
- GitHub Actions
Observability
- Prometheus
- Grafana
- Elasticsearch
- Kibana
Security
- Vault
- Trivy
- SonarQube
- Red Hat
Data & storage
- Cloudera
- Hadoop
- Spark
- Kafka
- Cassandra
- Ceph
- PostgreSQL
- Redis
Runtime & AI
- Linux
- Ubuntu
- NGINX
- Python
- Ollama
Certified
- ● Certified Kubernetes Administrator
- ● AWS Solutions Architect
- ● Azure Fundamentals
03 / About
Abdul Hadi, in short.
Abdul Hadi is a cloud, DevOps and AI infrastructure engineer based in Pakistan. He works as a Solutions Architect at SazebTech Consulting, where he builds Kubernetes and OpenShift platforms, runs cloud migrations on AWS and Azure, and stands up private cloud infrastructure: IBM Storage Ceph, highly available PostgreSQL with Patroni, Cloudera clusters and enterprise key management.
He is a Certified Kubernetes Administrator (CKA) and an AWS Certified Solutions Architect. His research, Secure but Broken?, tests whether LLM-written fixes for Kubernetes security misconfigurations still work once deployed. He writes hands-on runbooks on Medium and runs AWS and AI workshops for computer science students.
- Name
- Abdul Hadi
- Role
- Solutions Architect, SazebTech Consulting
- Focus
- Kubernetes and OpenShift platforms, AWS and Azure, private cloud, DevSecOps, LLM inference
- Certifications
- Certified Kubernetes Administrator (CKA), AWS Certified Solutions Architect, Azure Fundamentals
- Education
- BS Computer Science, University of Central Punjab
- Based in
- Pakistan (UTC+5), working with teams worldwide
04 / Deploy log
Where I've shipped, newest first.
- [now]
Solutions Architect · SazebTech Consulting
Cloud migrations, Kubernetes platforms and private cloud builds for enterprise clients: HA PostgreSQL with Patroni, Ceph object storage behind HAProxy and Keepalived, Thales CipherTrust key management.
- [2026]
Author, Secure but Broken? · Independent research
Deployment-validated evaluation of LLM security repairs for Kubernetes. Paper, protocol and data are public.
- [2026]
Workshop lead · AI on AWS for CS students
Taught students to ship a Bedrock agent on Lambda and DynamoDB from an empty AWS account.
- [cert]
CKA · AWS Solutions Architect · Azure Fundamentals · Certifications
Published a CKA practice question set and an exam guide on Medium.
- [2023]
Speed Coding winner · PUCON '23
Moved fast. Didn't break things.
- [edu]
BS Computer Science · University of Central Punjab
Started with React and React Native, ended up in infrastructure.
05 / Writing
Runbooks I wish I'd had.
- Connect GitHub Actions to Azure using OIDC (no passwords)
- From 2.5 t/s to 27 t/s: optimizing local LLMs on AWS CPU instances
- Your first CI/CD pipeline in 5 minutes: GitHub Actions to AWS Lightsail
- From 1,500 to 127,000 IOPS: our journey with AWS EKS storage
- High-availability Ceph Object Gateway ingress with HAProxy and Keepalived
- Deploying a highly available PostgreSQL cluster on Linux with Patroni
- Thales CipherTrust Manager on a private cloud
- IBM Storage Ceph installation guide