Sami U.

I take AI prototypes to production: MLOps and DevOps engineer for LLM apps, RAG and machine learning models on AWS, GCP and Azure. MS in AI, university MLOps lecturer, Top Rated. If your model works in a notebook but breaks with real users, real data or real cloud bills, that's the gap I close. I build the deployment, CI/CD, monitoring and infrastructure around your AI so it stays up, stays fast and stays affordable. RECENT WORK 1. CI/CD pipelines for a React Native product: GitHub Actions, automated builds and releases 2. Fine-tuned text-to-speech model served on AWS EC2 GPUs behind FastAPI, with GPU cost optimisation 3. Search relevance tuning in production on Google Cloud Vertex AI 4. Multi-tenant insurance AI platform: security hardening, CI/CD and test automation 5. Internal operations platform with an AI Slack assistant, used daily by 30 staff WHAT I DO 1. Deploy machine learning and LLM models to production: model serving and inference APIs with Docker, Kubernetes, FastAPI and serverless 2. CI/CD pipelines (continuous integration and delivery) with GitHub Actions, GitLab CI and Jenkins 3. Infrastructure as code with Terraform, Helm and ArgoCD 4. Productionise RAG apps and AI agents (LangChain, LangGraph, OpenAI API) with LLMOps practices 5. Monitoring and observability: Prometheus, Grafana, OpenTelemetry, drift detection 6. GPU and cloud cost optimisation on AWS (EC2, S3, Lambda), GCP (Vertex AI) and Azure 7. Data and ML pipelines: Airflow, DVC, MLflow, Kafka ROLES I FILL MLOps engineer, DevOps engineer, MLOps expert, AI infrastructure specialist, cloud architect, LLMOps consultant, platform engineer. One-off deployment or ongoing support. PROBLEMS CLIENTS BRING ME 1. We have an AI prototype, but need to deploy it properly. 2. Our ML model works locally, but not reliably in production. 3. Our LLM app or RAG system is slow, costly, or hard to monitor. 4. We need CI/CD, Docker and Kubernetes for our AI product. 5. Our AI chatbot/call center works in testing, but breaks under real user load or costs too much to run at scale. 6. Our ML pipeline breaks when new data arrives—we need Airflow, DVC or MLflow to manage it. 7. Our search results or RAG relevance has degraded, and we don't know why or how to measure it. 8. We built an AI writing/content tool, but it's stuck in MVP—we need production infrastructure. 9. We're losing money to GPU costs on AWS/GCP. Our inference is unoptimized. 10. Our team can't reproduce ML experiments or track what version of the model is live. 11. We don't have monitoring or alerts for model drift, so we ship degraded models without knowing. WHY WORK WITH ME - I teach MLOps, DevOps and LLMOps to final-year computer science students at top tech institute of the Pakistan and hold an MS in AI & Software Engineering. - Two-time winner of the NLBSE tool competition (2025, 2026), with published research on BERT-based code comment classification. - You get production engineering with the reasoning written down: documented architecture, tests, and a handover your team can run without me. - You work with me directly, from first call to handover. Need a full product team for web or mobile? I co-run an agency on Upwork for that. Message me and I'll point you to it. STACK AWS, GCP, Azure, Docker, Kubernetes, Helm, Terraform, ArgoCD, GitHub Actions, GitLab CI, Jenkins, Linux, Prometheus, Grafana, MLflow, Airflow, DVC, Weights & Biases, Kafka, Python, FastAPI, PostgreSQL, Redis, LangChain, LangGraph, Hugging Face, PyTorch, Node.js, Next.js Send me a message describing what you're trying to ship. I'll reply with how I'd deploy it and a fixed-price estimate.