MLOps · Model deployment · Madrid
MLOps in Madrid
I take machine learning models to production and keep them alive: an inference API with FastAPI, Docker containers, a version registry, drift monitoring with PSI and observability with Prometheus and Grafana. The model that works in the notebook also works in production.
The same pipeline that trains is the one that serves
Preprocessing lives inside the scikit-learn pipeline, so what happens at training time happens exactly the same way at inference time. There are no two versions of the same code that can drift apart.
Version registry and reproducibility
Every trained model is registered with its validation metrics, hyperparameters and data hash. You can go back to any previous version and know exactly what it was trained on. The pointer to the active version lives in the registry, not in code.
Drift monitoring before the business notices
A reference profile of the training data and Population Stability Index metrics over live traffic. When data changes, the system detects it and surfaces it in Grafana before the model starts failing silently.
Reproducible infrastructure with Docker
Multi-stage image: builder installs dependencies, runtime only copies the virtual environment and runs as a non-root user. The same container that passes tests is the one that reaches production. No environment surprises.
Have a model that needs to reach production?
Write to me with the current state: notebook, script, half-finished API, whatever. I will tell you what it takes to make it work in production and not break.
Start a projectFAQ
What is MLOps and why does it matter?
MLOps is the engineering practice that allows ML models to reach production and stay alive: versioning, CI/CD, monitoring, retraining. Without MLOps, models die in notebooks or degrade without anyone knowing. More and more companies have data and want to extract value from it, but most do not have engineers with experience in this.
What technical stack do you use for MLOps?
FastAPI for the inference API (Pydantic input validation, synchronous and batch endpoints), multi-stage Docker, Prometheus for metrics, Grafana for visualisation and PSI for drift. CI with GitHub Actions. For version registry, a lightweight file-based solution that needs no additional infrastructure.
Can I integrate MLOps with my existing infrastructure?
In most cases yes. If you already have Kubernetes, a cloud provider or tools like MLflow or Weights and Biases, the stack adapts. The priority is that deployment is reproducible and monitoring is running, not that you use a specific tool.
How much does setting up an MLOps platform cost?
A working MVP (inference API + Docker + basic monitoring) is between €2,000 and €4,000 depending on model complexity and infrastructure requirements. A complete platform with CI/CD, version registry, drift and Grafana dashboards is between €5,000 and €10,000. Call me for a concrete quote.
What happens when production data changes?
The system detects drift via PSI and surfaces it in Grafana. You will have visibility before the model starts giving bad results. Retraining can be automated or kept manual, depending on how frequently data changes and how much the team trusts the process.