diff --git a/tests/test_nlp.py b/tests/test_nlp.py
new file mode 100644
index 0000000..63b13f0
--- /dev/null
+++ b/tests/test_nlp.py
@@ -0,0 +1,35 @@
+import pytest
+import numpy as np
+from openrecall.nlp import cosine_similarity
+
+def test_cosine_similarity_identical_vectors():
+    a = np.array([1, 0, 0])
+    b = np.array([1, 0, 0])
+    assert cosine_similarity(a, b) == 1.0
+
+def test_cosine_similarity_orthogonal_vectors():
+    a = np.array([1, 0, 0])
+    b = np.array([0, 1, 0])
+    assert cosine_similarity(a, b) == 0.0
+
+def test_cosine_similarity_opposite_vectors():
+    a = np.array([1, 0, 0])
+    b = np.array([-1, 0, 0])
+    assert cosine_similarity(a, b) == -1.0
+
+def test_cosine_similarity_non_unit_vectors():
+    a = np.array([3, 0, 0])
+    b = np.array([1, 0, 0])
+    assert cosine_similarity(a, b) == 1.0
+
+def test_cosine_similarity_arbitrary_vectors():
+    a = np.array([1, 2, 3])
+    b = np.array([4, 5, 6])
+    expected_similarity = np.dot(a, b) / (np.linalg.norm(a) * np.linalg.norm(b))
+    assert cosine_similarity(a, b) == pytest.approx(expected_similarity)
+
+def test_cosine_similarity_zero_vector():
+    a = np.array([0, 0, 0])
+    b = np.array([1, 0, 0])
+    result = cosine_similarity(a, b)
+    assert np.isnan(result), "Expected result to be NaN when one of the vectors is a zero vector"