diff --git a/src/lightspeed_evaluation/core/embedding/ragas.py b/src/lightspeed_evaluation/core/embedding/ragas.py
index 0ad4c781..abfcd9d6 100644
--- a/src/lightspeed_evaluation/core/embedding/ragas.py
+++ b/src/lightspeed_evaluation/core/embedding/ragas.py
@@ -20,44 +20,45 @@ def __init__(self, embedding_manager: EmbeddingManager):
         Args:
             embedding_manager: Pre-configured EmbeddingManager with validated parameters
         """
-        config = embedding_manager.config
-        self.config = config
+        self.config = embedding_manager.config
 
         # Map provider names to litellm format
-        provider = config.provider.lower()
-        model = config.model
+        provider = self.config.provider.lower()
+        model = self.config.model
+        # Get additional provider kwargs
+        kwargs: dict[str, Any] = {}
+        if self.config.provider_kwargs:
+            kwargs.update(self.config.provider_kwargs)
 
-        # Build the model string for litellm
-        # Only OpenAI, Gemini, and HuggingFace are supported
-        if provider == "openai":
-            model_str = model  # OpenAI models don't need prefix
+        # Provider-specific configuration
+        if provider in ["openai", "gemini"]:
+            logger.debug("Using %s provider with model: %s", provider, model)
+            actual_provider = (
+                "litellm"  # Litellm provider auto-creates client in embedding_factory
+            )
         elif provider == "huggingface":
-            model_str = f"huggingface/{model}"
-        elif provider == "gemini":
-            model_str = f"gemini/{model}"
+            # HuggingFace default is use_api=False (local sentence-transformers)
+            # Only set explicitly if user hasn't overridden in provider_kwargs
+            kwargs.setdefault("use_api", False)
+            logger.debug(
+                "Using HuggingFace provider with model: %s (local=%s)",
+                model,
+                not kwargs["use_api"],
+            )
+            actual_provider = "huggingface"
         else:
-            logger.error("Unknown embedding provider: %s", config.provider)
-            raise ConfigurationError(f"Unknown embedding provider {config.provider}")
-
-        logger.debug(
-            "Using embedding provider: %s with model: %s -> %s",
-            provider,
-            model,
-            model_str,
-        )
-
-        # Get additional provider kwargs
-        kwargs: dict[str, Any] = {}
-        if config.provider_kwargs:
-            kwargs.update(config.provider_kwargs)
+            logger.error("Unknown embedding provider: %s", self.config.provider)
+            raise ConfigurationError(
+                f"Unknown embedding provider {self.config.provider}"
+            )
 
-        # Create embeddings using ragas 0.4+ embedding_factory with litellm
+        # Create embeddings using ragas 0.4+ embedding_factory
         # Cast to BaseRagasEmbedding as embedding_factory returns union type
         self.embeddings: BaseRagasEmbedding = cast(
             BaseRagasEmbedding,
             embedding_factory(
-                "litellm",
-                model=model_str,
+                provider=actual_provider,
+                model=model,
                 **kwargs,
             ),
         )
diff --git a/tests/unit/core/embedding/__init__.py b/tests/unit/core/embedding/__init__.py
new file mode 100644
index 00000000..0ca77f12
--- /dev/null
+++ b/tests/unit/core/embedding/__init__.py
@@ -0,0 +1 @@
+"""Unit tests for embedding module."""
diff --git a/tests/unit/core/embedding/conftest.py b/tests/unit/core/embedding/conftest.py
new file mode 100644
index 00000000..1967df35
--- /dev/null
+++ b/tests/unit/core/embedding/conftest.py
@@ -0,0 +1,10 @@
+"""Fixtures for embedding unit tests."""
+
+import pytest
+from pytest_mock import MockerFixture, MockType
+
+
+@pytest.fixture
+def mock_embedding_factory(mocker: MockerFixture) -> MockType:
+    """Mock embedding_factory for ragas embedding tests."""
+    return mocker.patch("lightspeed_evaluation.core.embedding.ragas.embedding_factory")
diff --git a/tests/unit/core/embedding/test_ragas.py b/tests/unit/core/embedding/test_ragas.py
new file mode 100644
index 00000000..077ca50c
--- /dev/null
+++ b/tests/unit/core/embedding/test_ragas.py
@@ -0,0 +1,71 @@
+"""Unit tests for RagasEmbeddingManager."""
+
+import pytest
+from pytest_mock import MockerFixture, MockType
+
+from lightspeed_evaluation.core.embedding.manager import EmbeddingManager
+from lightspeed_evaluation.core.embedding.ragas import RagasEmbeddingManager
+from lightspeed_evaluation.core.models import EmbeddingConfig
+
+
+class TestRagasEmbeddingManagerProviderRouting:
+    """Test provider routing logic in RagasEmbeddingManager."""
+
+    def test_huggingface_uses_native_provider(
+        self, mocker: MockerFixture, mock_embedding_factory: MockType
+    ) -> None:
+        """Verify HuggingFace uses native provider, not litellm.
+
+        HuggingFace embeddings should use the 'huggingface' provider directly
+        to support local sentence-transformers models with use_api=False.
+        """
+        mocker.patch(
+            "lightspeed_evaluation.core.embedding.manager.check_huggingface_available"
+        )
+
+        config = EmbeddingConfig(
+            provider="huggingface",
+            model="sentence-transformers/all-MiniLM-L6-v2",
+        )
+        embedding_manager = EmbeddingManager(config)
+
+        RagasEmbeddingManager(embedding_manager)
+
+        mock_embedding_factory.assert_called_once_with(
+            provider="huggingface",
+            model="sentence-transformers/all-MiniLM-L6-v2",
+            use_api=False,
+        )
+
+    @pytest.mark.parametrize(
+        "provider,model",
+        [
+            ("openai", "text-embedding-3-small"),
+            ("gemini", "text-embedding-004"),
+        ],
+    )
+    def test_cloud_providers_use_litellm(
+        self,
+        mocker: MockerFixture,
+        mock_embedding_factory: MockType,
+        provider: str,
+        model: str,
+    ) -> None:
+        """Verify OpenAI and Gemini use litellm provider.
+
+        Cloud providers (OpenAI, Gemini) should route through 'litellm'
+        which auto-creates the appropriate client in embedding_factory.
+        """
+        mocker.patch(
+            "lightspeed_evaluation.core.embedding.manager.validate_provider_env"
+        )
+
+        config = EmbeddingConfig(provider=provider, model=model)
+        embedding_manager = EmbeddingManager(config)
+
+        RagasEmbeddingManager(embedding_manager)
+
+        mock_embedding_factory.assert_called_once_with(
+            provider="litellm",
+            model=model,
+        )