diff --git a/.github/workflows/pyright.yml b/.github/workflows/pyright.yml
new file mode 100644
index 0000000..8150b0b
--- /dev/null
+++ b/.github/workflows/pyright.yml
@@ -0,0 +1,27 @@
+name: Pyright Type Checking
+
+on:
+  push:
+    branches: [ master ]
+  pull_request:
+    branches: [ master ]
+
+jobs:
+  pyright:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: '3.11'
+
+      - name: Install dependencies
+        run: |
+          pip install .[dev]
+
+      - name: Run pyright
+        run: pyright
+        env:
+          FORCE_COLOR: "1"
diff --git a/preprocessing_utils.py b/preprocessing_utils.py
index 437db69..d68045b 100644
--- a/preprocessing_utils.py
+++ b/preprocessing_utils.py
@@ -32,13 +32,14 @@ def create_val_set(csv_file, val_fraction):
     out of it specified by val_fraction.
     """
     csv_file = Path(csv_file)
-    dataset = pd.read_csv(csv_file)
+    dataset: pd.DataFrame = pd.read_csv(csv_file)
     np.random.seed(0)
     dataset_mod = dataset[dataset.toxic != -1]
     indices = np.random.rand(len(dataset_mod)) > val_fraction
     val_set = dataset_mod[~indices]
     output_file = csv_file.parent / "val.csv"
     logger.info("Validation set saved to %s", output_file)
+    assert isinstance(val_set, (pd.DataFrame, pd.Series))
     val_set.to_csv(output_file)
 
 
diff --git a/pyproject.toml b/pyproject.toml
index 09033ad..9325a86 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -33,7 +33,8 @@ dev = [
     "scikit-learn >= 0.23.2",
     "tqdm",
     "pre-commit",
-    "numpy>=2"
+    "numpy>=2",
+    "pyright"
 ]
 
 [tool.ruff]
diff --git a/run_prediction.py b/run_prediction.py
index 2871e44..df22066 100644
--- a/run_prediction.py
+++ b/run_prediction.py
@@ -32,7 +32,10 @@ def run(model_name, input_obj, dest_file, from_ckpt, device="cpu"):
         model = Detoxify(checkpoint=from_ckpt, device=device)
     res = model.predict(text)
 
-    res_df = pd.DataFrame(res, index=[text] if isinstance(text, str) else text).round(5)
+    res_df = pd.DataFrame(
+        res,
+        index=[text] if isinstance(text, str) else text,  # pyright: ignore[reportArgumentType]
+    ).round(5)
     print(res_df)
     if dest_file is not None:
         res_df.index.name = "input_text"
diff --git a/src/data_loaders.py b/src/data_loaders.py
index 2f58741..8796012 100644
--- a/src/data_loaders.py
+++ b/src/data_loaders.py
@@ -49,6 +49,8 @@ def load_data(self, csv_file):
             filtered_change_names = {k: v for k, v in change_names.items() if k in final_df.columns}
             if len(filtered_change_names) > 0:
                 final_df.rename(columns=filtered_change_names, inplace=True)
+        else:
+            raise TypeError("Invalid input type for csv_file, must be a string or a list of strings")
         return final_df
 
     def load_val(self, test_csv_file, add_labels=False):
@@ -155,6 +157,8 @@ def __getitem__(self, index):
         meta["text_id"] = text_id
 
         if self.train:
+            if self.weights is None:
+                raise Exception("self.weights must not be None")
             meta["weights"] = self.weights[index]
             toxic_weight = self.weights[index] * self.loss_weight * 1.0 / len(self.classes)
             identity_weight = (1 - self.loss_weight) * 1.0 / len(self.identity_classes)