Aafaq Rana, Senior DevOps Engineer and SRE

Cloud Migration & Security Engineer (contract), Independent Cloud & DevOps Consultant

Oct 2026 to Present

Migrating a client platform to Kubernetes and locking down its edge, from the change plan to the cutover.

• Cloudflare edge security and DNS: DNS management, WAF rules, bot management and bot blocking, rate limiting and TLS, hardening sites and applications against abuse and attack traffic.
• DigitalOcean migration from Droplets to DigitalOcean Kubernetes (DOKS) with a written infrastructure change plan, target architecture, phased cutover and rollback, provisioned with infrastructure as code and secured by default.
• Sentry error and performance monitoring across client applications, plus hardening of servers, containers and network exposure as part of the migration.

Stack: Cloudflare, DigitalOcean, DOKS, Terraform, Sentry

Senior DevOps Engineer, DevOps Team Lead, Netsol Technologies

Dec 2021 to Present

I lead the DevOps team behind Netsol's asset finance and leasing software for the automotive and financing industry, across AWS, Azure, GCP and OCI, from GPU inference on EKS to clusters with no internet at all.

• Moved delivery off Jenkins and SVN onto standardised pipelines. Build and ship time dropped by about half, delivery throughput rose by about a third.
• Run Kubernetes from managed EKS to fully air-gapped K3s on customer bare metal, with every image, chart and operator mirrored.
• Serve LLMs on dedicated GPU node groups with vLLM, and alert on real GPU utilisation from DCGM metrics instead of CPU proxies.
• Owned the infrastructure control set through a SOC 2 audit: IAM and KMS policy, secrets, network segmentation and evidence.
• Built a Kubernetes agent on the MCP server with LangGraph that watches cluster health and catches failures before users do.
• Turned observability pipelines into agentic workflows with LangChain and LangGraph: AI incident triage, automated root-cause analysis and self-healing across Prometheus, Grafana and EFK.
• Documented legacy codebases automatically into a RAG knowledge base on ChromaDB (LangChain, LlamaIndex), cutting onboarding and incident time and lifting DORA metrics.
• Put platforms behind Cloudflare (DNS, WAF, bot blocking, rate limiting, TLS) and moved workloads from DigitalOcean Droplets to DOKS with a change plan, phased cutover and rollback, all in IaC, with Sentry watching the apps.
• Run Karpenter node pools with Spot and consolidation, and KEDA on external signals, so compute tracks real demand.

Stack: EKS, K3s, Terraform, Pulumi, ArgoCD, vLLM, LangGraph, Karpenter, Cloudflare, Sentry, SOC 2

Sr. DevOps, AI Trainer and Evaluator (contract), Turing

Jun 2025 to Jul 2026

Pod lead across two contracts: infrastructure-as-code evaluation and model training for a major cloud provider, then computer-use and company benchmarks for AI agents.

• IaC contract: led the pod building real infrastructure with Terraform, CDK, CloudFormation and Pulumi, and grading model-written IaC with RLHF, preference ranking, reward modelling, pass@k and HumanEval.
• Built the Test Automation Platform (TAP) with Terraform, CDKTF, AWS CDK, CloudFormation and Pulumi to provision isolated, encrypted AWS and DigitalOcean environments for model training and evaluation.
• Designed adversarial multi-step agentic scenarios on Kubernetes, Terraform and cloud architecture, and wrote the preference data and rubrics that improved IaC code generation.
• Benchmark contract: led the pod building computer-use and company workflow tasks for agents, each in a sandboxed container with an oracle solution, a binary grader and difficulty runs across several frontier models.
• Set the quality bar for both pods: reviews against a 14-dimension rubric, difficulty bands, and the runbook from claim to ready for review.
• Wrote Python tooling for packaging, verification and batch runs, and read failure patterns in agent trajectories.

Stack: Terraform, CDKTF, Pulumi, RLHF, pass@k, HumanEval, Docker, Python

Infrastructure Benchmark Engineer (contract), Vera Platforms

Jul 2026 to Sep 2026

Terraform tasks across AWS, Cloudflare and DigitalOcean, built to make frontier agents fail, graded against live cloud state.

• Setup provisions real AWS resources and seeds Terraform state into the agent sandbox, with the answer key kept out of reach.
• Graders check the live end state with real API calls as different identities, never the agent's own report.

Stack: Terraform, AWS, Cloudflare, DigitalOcean, boto3

Associate Software Engineer, Devsinc

Jul 2020 to Dec 2021

Where I learned Linux properly and shipped my first production pipelines.

• Deployed Python FastAPI services to AWS and DigitalOcean Droplets through Azure DevOps pipelines, behind Cloudflare DNS and TLS, with Sentry error tracking and infrastructure in Terraform.
• Built a document processing application that ran about 20% faster than the system it replaced.

Stack: FastAPI, Terraform, Azure DevOps, AWS, DigitalOcean

Resilient 3-tier platform on AWS

Web, API and Postgres on ECS Fargate, shipped through a security-gated pipeline and proven with chaos tests.

Eight CDK stacks: WAF and CloudFront in front, a load balancer that only accepts CloudFront, Fargate tiers across two zones and Multi-AZ Postgres in subnets with no route out. Four Azure DevOps pipelines gate every release with SAST, SCA, secret, IaC and DAST scans, deploy with zero downtime, then kill tasks under live load and pass only on zero failed requests.

Industrial LLM platform

Serving, benchmarking and operating LLMs and vision models on open-source tooling.

Qwen2.5 served with vLLM, TensorRT-LLM and llama.cpp behind a reproducible benchmark harness on free Kaggle T4s: AWQ INT4 more than doubled single-user speed and hit 908 tok/s at 32 users, 1,121 tok/s with tensor parallelism on two GPUs. YOLOv8n on TensorRT FP16 ran at 295 images/s with no mAP loss, served through Triton. A cited Haystack RAG API over Elasticsearch hybrid retrieval reaches recall@5 of 0.87. OpenTelemetry, Prometheus, Loki and Langfuse feed SLO burn-rate alerts; Helm, ArgoCD, KEDA, Jenkins and GitHub Actions ship it to Kubernetes and OpenShift. Every number traces to a raw log.

https://github.com/aafaq01/industrial-llm-stack

Fathom rebuild

An AI meeting notetaker, built and shipped in 24 hours.

Diarized transcripts, owner-grouped action items and clip pages, built for the long, crowded call. The model cites transcript rows and never writes a timestamp, so times cannot drift.

https://github.com/aafaq01/fathom-rebuild

Obsly

Observability where errors, traces and releases share one key.

Own wire protocol, own SDKs, own ingest. One click walks from a browser event to the backend span, the SQL query and the commit.

https://github.com/aafaq01/obsly

Job Platform

Local-first job application automation.

Paste a job link, get a tailored resume PDF and a tracked application. FastAPI and LangGraph, quality gates before any deploy.

https://github.com/aafaq01/job-platform

Kubernetes failure agent

An MCP agent that catches cluster trouble early.

Built on the Kubernetes MCP server with LangGraph. Watches health, predicts failures and drives remediation with kubectl-native tools.

GPU inference on EKS

LLM serving that alerts on the metric users feel.

Dedicated GPU node groups, NVIDIA device plugin, vLLM serving and DCGM metrics in Prometheus.

MLOps infrastructure automation

Model deployment that provisions its own environment and serves securely.

Terraform, ECS Fargate, S3 and CI/CD for automated model deployment, per-environment provisioning and secure inference for AI applications.

App Runner platform

A document service behind CloudFront and WAF, all in Terraform.

Two-phase deploy: CI/CD pipeline first, runtime second, with rate limiting, secrets and logs wired in.

https://github.com/aafaq01/app-runner-deployment

Air-gapped Kubernetes for regulated customers

Clusters on customer bare metal with no internet at all.

Every image, chart and operator mirrored, PostgreSQL operators included, packaged so on-site staff who are not Kubernetes engineers can install and run it.