Fayaz Abdul

Worked on a public sector digital transformation project where we migrated around 1,400 servers to the public cloud. Started with a lift and shift approach, but in the process we automated every aspect from creating images using Packer, creating infrastructure using Terraform, coding Ansible playbooks for configuring the servers, writing Jenkins groovy pipelines for CI/CD, storing the secrets in vault, using consul for service discovery, moving to serverless approach where feasible, converting monolithic applications to microservices, moved from pet to cattle model, containerization using Docker, Docker-Compose, using container orchestration tools such as Rancher, Docker Swarm, Kubernetes, and implemented and improved monitoring tools such as ELK, Zabbix, and Prometheus.

Architected, designed, and implemented the application, tooling, monitoring layers from infra and DevOps perspective to support the big data applications to be redundant, highly available, and scalable with blue-green/zero downtime deployments. As an end result, migrated around 21 environments successfully without compromising on security, adhering to CIS standards, and passing all ITHC.

Did many upgrades and POCs on hybrid cloud, PCF, and private clouds.

Created the Azure Kubernetes Service along with the whole eco-system such as resource groups, virtual networks, storage accounts, etc.., using Terraform and used Terragrunt to reuse the same code to deploy for different environments.

Setup the Azure DevOps build pipelines and release pipelines which will get triggered on developer commits and build the required docker images and push the same to Azure container registry with appropriate tags, generate the artifacts and pushed the same as part of rolling updates in release pipelines.

Enabled the logging, and used different patterns of deployment in Kubernetes as this is for AI/ML application. Used node pools and enabled autoscaling for the same and made sure for ML GPU workloads only one pod runs on a given node(with Nvidia drivers installed as part of Daemonset) and based on the resource limits spin up a new node for a new ML GPU workload pod.

Architected and implemented a low latency trading platform that can execute a trading order in all the client accounts at the same time under 500ms. Used Kafka with many consumers one per user and single or limited producers. Complexity is to be able to make it compatible for all different types of accounts, designing an order management system.