civo · Rishit-dagli · Oct 30, 2023 · Dec 16, 2023 · Dec 23, 2023 · kaihoffman
diff --git a/content/docs/machine-learning/images/mnist-pipeline.png b/content/docs/machine-learning/images/mnist-pipeline.png
diff --git a/content/docs/machine-learning/quickstart.md b/content/docs/machine-learning/quickstart.md
@@ -0,0 +1,73 @@
+---
+sidebar_position: 2
+title: Get Started with Kubeflow
+description: Learn how to get started with Civo's Kubeflow via an example.
+---
+
+<head>
+  <title>Kubeflow quickstart | Civo Documentation</title>
+</head>
+
+In this quickstart, focused mainly on Pipelines and Notebooks, we take a look at an example to train and run a machine learning model using Kubeflow.
+
+## Before you start
+
+Make sure you can log into your Kubeflow cluster by following the instructions at [Logging into your cluster](kubeflow-dashboard.md#logging-into-your-cluster).
+
+Make sure you can create a notebook and log into the notebook by following the instructions at [Creating a Kubeflow Notebook](creating-a-new-kubeflow-notebook.md).
+
+## Run a pipeline
+
+We will first compile a pipeline DSL that will train a model using the [MNIST dataset](https://ieeexplore.ieee.org/document/726791). Notice that this pipeline mainly has the following coponents which are run: hyperparameter tuning with Katib, creating a new volume for training, run a training job and finally serve the model using KServe.
+
+Now [create a new notebook instance](creating-a-new-kubeflow-notebook.md) and run the following commands:
+
+```bash
+git clone 
+cd 
+pip install -r requirements.txt
+python mnist-example.py
+```
+
+This produces for us a `mnist-example.yaml` file that we will use to run our pipeline.
+
+We now use [Pipelines](kubeflow-dashboard.md/) and use the pipeline configuration we generated in the previous step to define the pipeline. Once you do so, you should create a new [experiment](kubeflow-dashboard.md/) and then trigger a run for the pipeline by going to the Run tab, clicking on the "create a run" button and choosing the pipeline you just created.
+
+<img src={require('./images/mnist-pipeline.png').default} style={{"background-color":"white"}} />
+
+This would trigger a Kubeflow pipeline run and we would see the pipeline run in the Pipelines dashboard. Once the pipeline run is complete, we can see the model saved in the `end-to-end-pipeline-{ID}-model` volume we created earlier.
+
+## Predict using the model
+
+We can now use the model we trained to make predictions. We will use the [KFServing](kubeflow-dashboard.md/) component of Kubeflow to serve our model. You could now modify this code and run it in the Notebook instance you created to predict using the Serving endpoint.
+
+```python
+import numpy as np
+from PIL import Image
+import requests
+
+name = "kfaas-docs"
+namesapce = "my-profile"
+image_url = "https://raw.githubusercontent.com/kubeflow/katib/master/examples/v1beta1/kubeflow-pipelines/images/9.bmp"
+image = Image.open(requests.get(image_url, stream=True).raw)
+data = (
+    np.array(image.convert("L").resize((28, 28)))
+    .astype(np.float)
+    .reshape(-1, 28, 28, 1)
+)
+data_formatted = np.array2string(
+    data, separator=",", formatter={"float": lambda x: "%.1f" % x}
+)
+json_request = '{{ "instances" : {} }}'.format(data_formatted)
+
+url = "http://{}-predictor-default.{}.svc.cluster.local/v1/models/{}:predict".format(
+    name, namespace, name
+)
+response = requests.post(url, data=json_request)
+
+print("Prediction for the image")
+display(image)
+print(response.json())
+```
+
+You should see the image and the JSON response our prediction endpoint returns for the image.