diff --git a/.dstack/workflows/conda.yaml b/.dstack/workflows/conda.yaml
index a95d960..a8448e1 100644
--- a/.dstack/workflows/conda.yaml
+++ b/.dstack/workflows/conda.yaml
@@ -21,3 +21,14 @@ workflows:
     commands:
       - conda activate myenv
       - python usage/conda/hello_pandas.py
+
+  - name: xgboost-env
+    help: "This workflow prepares myenv Conde environment with xgboost installed."
+    commands:
+      - conda env create --file usage/conda/xgboost.yaml
+      - conda activate xgboost-env
+      - conda install -y scikit-learn
+      - conda install -y xgboost
+    provider: bash
+    artifacts:
+      - path: /opt/conda/envs/xgboost-env
diff --git a/.dstack/workflows/xgboost.yaml b/.dstack/workflows/xgboost.yaml
new file mode 100644
index 0000000..48b56f9
--- /dev/null
+++ b/.dstack/workflows/xgboost.yaml
@@ -0,0 +1,8 @@
+workflows:
+  - name: xgboost
+    provider: bash
+    deps:
+      - workflow: xgboost-env
+    commands:
+      - conda activate xgboost-env
+      - python examples/xgboost/train.py ${{ run.args }}
diff --git a/examples/xgboost/train.py b/examples/xgboost/train.py
new file mode 100644
index 0000000..1b12330
--- /dev/null
+++ b/examples/xgboost/train.py
@@ -0,0 +1,48 @@
+import argparse
+
+from sklearn.datasets import load_iris
+from sklearn import metrics, model_selection
+import xgboost as xgb
+
+
+def parse_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--learning-rate",
+        default=0.2,
+        type=float
+    )
+    return vars(parser.parse_args())
+
+
+def main():
+    args = parse_args()
+    print(args)
+
+    iris = load_iris()
+    data, labels = iris.data, iris.target
+    labels = iris.target
+    data_train, data_test, labels_train, labels_test = model_selection.train_test_split(
+        data, labels, test_size=0.1, random_state=2023)
+
+    data_train = xgb.DMatrix(data_train, label=labels_train)
+    data_test = xgb.DMatrix(data_test, label=labels_test)
+    params = {
+        "learning_rate": args["learning_rate"],
+        "objective": "multi:softprob",
+        "seed": 2023,
+        "num_class": 3,
+    }
+    model = xgb.train(params, data_train, evals=[(data_train, "train")])
+
+    y_proba = model.predict(data_test)
+    y_pred = y_proba.argmax(axis=1)
+    loss = metrics.log_loss(labels_test, y_proba)
+    acc = metrics.accuracy_score(labels_test, y_pred)
+
+    print(f"Model trained: loss={loss:.2f}, acc={acc:.2f}")
+    
+
+if __name__ == "__main__":
+    main()
+
diff --git a/usage/conda/xgboost.yaml b/usage/conda/xgboost.yaml
new file mode 100644
index 0000000..e310987
--- /dev/null
+++ b/usage/conda/xgboost.yaml
@@ -0,0 +1,4 @@
+name: xgboost-env
+
+dependencies:
+  - python=3.10