diff --git a/.github/workflows/check_conflicts.yml b/.github/workflows/check_conflicts.yml
new file mode 100644
index 0000000..c5eeefb
--- /dev/null
+++ b/.github/workflows/check_conflicts.yml
@@ -0,0 +1,26 @@
+name: Check Conflicts
+
+on:
+  pull_request:
+    types: [opened, synchronize, reopened, edited]
+
+jobs:
+  check_conflicts:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v2
+
+      - name: Set up Python
+        uses: actions/setup-python@v3
+        with:
+          python-version: 3.9
+
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          python -m pip install pandas==2.2.2 PyYAML==6.0.1
+
+      - name: Run check_conflicts.py
+        run: python check_conflicts.py
diff --git a/check_conflicts.py b/check_conflicts.py
new file mode 100644
index 0000000..3701cf7
--- /dev/null
+++ b/check_conflicts.py
@@ -0,0 +1,60 @@
+import yaml
+import pandas as pd
+
+
+def check_conflicts(file_path):
+    with open(file_path, "r") as file:
+        data = yaml.safe_load(file)
+    df = pd.DataFrame(data["deployment"])
+    df = df.T
+    df["max_count"] = df["flavor"].apply(lambda x: data["nodes_inventory"][x])
+    df["node"] = df.index
+
+    training_df = df[df["node"].str.startswith("training-")]
+    if not training_df.empty:
+        training_df = training_df.reset_index()
+        training_df["date"] = training_df.apply(
+            lambda x: pd.date_range(start=x["start"], end=x["end"]), axis=1
+        )
+        training_df = training_df.explode("date")
+        training_df = training_df.groupby(["flavor", "date"]).agg(
+            {"count": "sum", "max_count": "first", "node": "unique"}
+        )
+        training_df = training_df.reset_index()
+        training_df["conflict"] = (
+            training_df["count"] > training_df["max_count"]
+        )
+
+    not_training_df = df[~df["node"].str.startswith("training-")]
+    not_training_df = not_training_df.groupby("flavor").agg(
+        {"count": "sum", "max_count": "first", "node": "unique"}
+    )
+    not_training_df = not_training_df.reset_index()
+    not_training_df["conflict"] = (
+        not_training_df["count"] > not_training_df["max_count"]
+    )
+    
+    df = pd.concat([training_df, not_training_df]) if not training_df.empty and not not_training_df.empty else training_df if not training_df.empty else not_training_df
+
+    df["node"] = df["node"].apply(lambda x: ", ".join(x))
+
+    conflicts = []
+    for _, row in df[df["conflict"]].iterrows():
+        date = (
+            f"on {row['date'].strftime('%Y-%m-%d')} "
+            if not pd.isnull(row["date"])
+            else ""
+        )
+        conflicts.append(
+            f"Conflict for {row['flavor']} {date}with {row['count']} nodes requested but only {row['max_count']} available. Conflicted nodes: {row['node']}."
+        )
+
+    conflicts = "\n".join(conflicts)
+    if conflicts:
+        raise ValueError(f"Conflicts found:\n{conflicts}")
+    else:
+        return True
+
+
+if __name__ == "__main__":
+    check_conflicts("resources.yaml")