diff --git a/README.md b/README.md
index 0185f93..7ede4e7 100644
--- a/README.md
+++ b/README.md
@@ -71,6 +71,10 @@ A collection of delicious docker recipes.
 - [x] hubot :octocat:
 - [x] jenkins-arm :beetle:
 
+## Machine Learning
+
+- [x] ludwig
+
 ## Cluster
 
 - [x] ggr
diff --git a/ludwig/Dockerfile b/ludwig/Dockerfile
new file mode 100644
index 0000000..c2daee9
--- /dev/null
+++ b/ludwig/Dockerfile
@@ -0,0 +1,14 @@
+FROM tensorflow/tensorflow:latest-py3
+
+RUN apt-get -y install git
+
+RUN git clone --depth=1 https://github.com/uber/ludwig.git \
+    && cd ludwig/ \
+    && pip install -r requirements.txt -r requirements_text.txt \
+          -r requirements_image.txt -r requirements_audio.txt \
+          -r requirements_serve.txt -r requirements_viz.txt \
+    && python setup.py install
+
+WORKDIR /data
+
+ENTRYPOINT ["ludwig"]
diff --git a/ludwig/README.md b/ludwig/README.md
new file mode 100644
index 0000000..8d28555
--- /dev/null
+++ b/ludwig/README.md
@@ -0,0 +1,70 @@
+ludwig
+======
+
+[Ludwig][1] is a toolbox that allows to train and test deep learning models
+without the need to write code.
+
+## up and running
+
+```bash
+$ mkdir -p data
+$ vim data/model.yaml
+
+$ wget http://boston.lti.cs.cmu.edu/classes/95-865-K/HW/HW2/epinions.zip
+$ unzip epinions.zip
+$ mv epinions/epinions-1.csv data/train.csv
+$ mv epinions/epinions-2.csv data/predict.csv
+
+$ tree data
+├── model.yaml
+├── predict.csv
+└── train.csv
+
+$ docker-compose run --rm train
+$ docker-compose run --rm visualize
+$ docker-compose run --rm predict
+$ docker-compose up -d serve
+
+$ curl http://127.0.0.1:8000/predict -X POST -F 'text=taking photos and recording videos'
+{
+  "class_predictions": "Camera",
+  "class_probabilities_<UNK>": 9.438252263072044e-11,
+  "class_probabilities_Auto": 0.32920214533805847,
+  "class_probabilities_Camera": 0.6707978248596191,
+  "class_probability": 0.6707978248596191
+}
+
+$ curl http://127.0.0.1:8000/predict -X POST -F 'text=looking to buy a new sports car'
+{
+  "class_predictions": "Auto",
+  "class_probabilities_<UNK>": 1.900043131457165e-15,
+  "class_probabilities_Auto": 0.9999126195907593,
+  "class_probabilities_Camera": 8.738834003452212e-05,
+  "class_probability": 0.9999126195907593
+}
+
+$ tree -L 3 data
+├── model.yaml
+├── predict.csv
+├── train.csv
+├── results
+│   └── experiment_run
+│       ├── description.json
+│       ├── model
+│       └── training_statistics.json
+├── results_0
+│   ├── class_predictions.csv
+│   ├── class_predictions.npy
+│   ├── class_probabilities.csv
+│   ├── class_probabilities.npy
+│   ├── class_probability.csv
+│   └── class_probability.npy
+└── visualize
+    ├── learning_curves_class_accuracy.png
+    ├── learning_curves_class_hits_at_k.png
+    ├── learning_curves_class_loss.png
+    ├── learning_curves_combined_accuracy.png
+    └── learning_curves_combined_loss.png
+```
+
+[1]: https://uber.github.io/ludwig/
diff --git a/ludwig/data/model.yaml b/ludwig/data/model.yaml
new file mode 100644
index 0000000..f129f9f
--- /dev/null
+++ b/ludwig/data/model.yaml
@@ -0,0 +1,11 @@
+input_features:
+    -
+        name: text
+        type: text
+        level: word
+        encoder: parallel_cnn
+
+output_features:
+    -
+        name: class
+        type: category
diff --git a/ludwig/docker-compose.yml b/ludwig/docker-compose.yml
new file mode 100644
index 0000000..5b08239
--- /dev/null
+++ b/ludwig/docker-compose.yml
@@ -0,0 +1,26 @@
+train:
+  image: vimagick/ludwig
+  command: train --data_csv train.csv -mdf model.yaml
+  volumes:
+    - ./data:/data
+
+visualize:
+  image: vimagick/ludwig
+  command: visualize -v learning_curves -trs results/experiment_run/training_statistics.json -od visualize -ff png
+  volumes:
+    - ./data:/data
+
+predict:
+  image: vimagick/ludwig
+  command: predict --data_csv predict.csv -m results/experiment_run/model
+  volumes:
+    - ./data:/data
+
+serve:
+  image: vimagick/ludwig
+  command: serve -m results/experiment_run/model -p 8000
+  ports:
+    - "8000:8000"
+  volumes:
+    - ./data:/data
+  restart: unless-stopped