Gokul Upadhyay Guragain

Delivery and MLOps work on a healthcare platform operating under HITRUST, where the pipeline is an audited control rather than convenience tooling.

• Automated infrastructure and deployments with Ansible, replacing manual host configuration with idempotent, version-controlled playbooks.
• Enabled MLOps workflows on AWS SageMaker so training became a reproducible pipeline execution and the model registry became the only route to an endpoint.
• Built and maintained secure GitLab CI pipelines for HITRUST-compliant healthcare systems, with each control stage emitting retained evidence rather than a transient log line.
• Strengthened monitoring, logging and security practices across ISO-aligned, data-sensitive environments.

Containerised ML pipelines and K3s-hosted inference, instrumented so that a model service is treated as a production service with an extra failure mode.

• Deployed inference services on K3s, giving training and serving identical environments through a shared, version-pinned base image.
• Applied monitoring and logging that pairs request metrics with prediction-distribution metrics, which is what surfaces a model that is healthy, fast and wrong.
• Built containerised ML pipelines with FastAPI and Docker for automated training and inference workflows.
• Implemented model versioning, experiment tracking and GitHub Actions CI/CD so a deployment and its model roll back together as one image digest.

AWS architecture work where availability, cost and blast radius were treated as one conversation, and Terraform was the primary artefact rather than an afterthought.

• Designed highly available AWS architectures across EC2, VPC, IAM, RDS, S3 and Application Load Balancer, with availability decided per failure domain and each answer carrying a stated cost.
• Automated infrastructure provisioning with Terraform, structured as composable modules so an environment difference is a variable file.
• Implemented IAM least-privilege policies derived from the workload rather than from service-level wildcards, plus network segmentation that removed flat internal connectivity.
• Applied cost optimisation through right-sizing against observed utilisation and scheduling non-production environments outside working hours.

Structured cloud engineering training built around practical tasks rather than lecture material.

• Worked through cloud engineering tasks against real-world use cases rather than isolated exercises.
• Built foundational skills in cloud architecture, deployment and infrastructure management that the Adex apprenticeship then put under deadline pressure.