diff --git a/pyproject.toml b/pyproject.toml
index 4ccbcee..474162b 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -30,8 +30,8 @@ dev = [
 
 [tool.basedpyright]
 reportAny = false
-reportExplicitAny = false
 enableTypeIgnoreComments = true
+reportUnreachable = false
 
 [tool.pytest.ini_options]
 asyncio_mode = "auto"
diff --git a/src/vector_search_mcp/engine.py b/src/vector_search_mcp/engine.py
deleted file mode 100644
index dabcc18..0000000
--- a/src/vector_search_mcp/engine.py
+++ /dev/null
@@ -1,40 +0,0 @@
-from collections.abc import Sequence
-from typing import Any, final
-
-from qdrant_client import AsyncQdrantClient, models
-
-from .config import Settings
-from .models import SearchRow
-
-
-@final
-class QdrantEngine:
-    def __init__(self) -> None:
-        self.settings = Settings()  # type: ignore[reportCallIssue]
-        self.client = AsyncQdrantClient(
-            url=self.settings.url, api_key=self.settings.api_key
-        )
-
-    async def semantic_search(
-        self,
-        embedding: Sequence[float] | models.NamedVector,
-        collection: str,
-        limit: int = 10,
-        conditions: Any | None = None,
-        threshold: float | None = None,
-    ) -> list[SearchRow]:
-        points = await self.client.search(
-            collection_name=collection,
-            query_vector=embedding,
-            query_filter=conditions,
-            limit=limit,
-            with_payload=True,
-            with_vectors=False,
-            score_threshold=threshold,
-        )
-
-        return [
-            SearchRow(chunk_id=str(point.id), score=point.score, payload=point.payload)
-            for point in points
-            if point.payload is not None
-        ]
diff --git a/src/vector_search_mcp/engine/__init__.py b/src/vector_search_mcp/engine/__init__.py
new file mode 100644
index 0000000..9dbf918
--- /dev/null
+++ b/src/vector_search_mcp/engine/__init__.py
@@ -0,0 +1,26 @@
+from enum import StrEnum
+from typing import Literal, overload
+
+from .qdrant_engine import QdrantEngine
+
+
+class EngineType(StrEnum):
+    QDRANT = "qdrant"
+    COSMOS = "cosmos"
+
+
+@overload
+def get_engine(backend: Literal[EngineType.QDRANT]) -> QdrantEngine: ...
+
+
+@overload
+def get_engine(backend: Literal[EngineType.COSMOS]) -> QdrantEngine: ...
+
+
+def get_engine(backend: EngineType):
+    if backend == EngineType.QDRANT:
+        return QdrantEngine()
+    elif backend == EngineType.COSMOS:
+        raise NotImplementedError("Cosmos engine is not implemented yet")
+    else:
+        raise ValueError(f"Unknown engine type: {backend}")
diff --git a/src/vector_search_mcp/engine/base_engine.py b/src/vector_search_mcp/engine/base_engine.py
new file mode 100644
index 0000000..0cb2b3e
--- /dev/null
+++ b/src/vector_search_mcp/engine/base_engine.py
@@ -0,0 +1,43 @@
+from abc import ABC, abstractmethod
+from typing import Generic, TypeVar
+
+from ..models import Condition, SearchRow
+
+ResponseType = TypeVar("ResponseType")
+ConditionType = TypeVar("ConditionType")
+
+__all__ = ["BaseEngine"]
+
+
+class BaseEngine(ABC, Generic[ResponseType, ConditionType]):
+    @abstractmethod
+    def transform_conditions(
+        self, conditions: list[Condition] | None
+    ) -> ConditionType | None: ...
+
+    @abstractmethod
+    def transform_response(self, response: ResponseType) -> list[SearchRow]: ...
+
+    @abstractmethod
+    async def run_similarity_query(
+        self,
+        embedding: list[float],
+        collection: str,
+        limit: int = 10,
+        conditions: ConditionType | None = None,
+        threshold: float | None = None,
+    ) -> ResponseType: ...
+
+    async def semantic_search(
+        self,
+        vector: list[float],
+        collection: str,
+        limit: int = 10,
+        conditions: list[Condition] | None = None,
+        threshold: float | None = None,
+    ) -> list[SearchRow]:
+        transformed_conditions = self.transform_conditions(conditions)
+        response = await self.run_similarity_query(
+            vector, collection, limit, transformed_conditions, threshold
+        )
+        return self.transform_response(response)
diff --git a/src/vector_search_mcp/engine/qdrant_engine.py b/src/vector_search_mcp/engine/qdrant_engine.py
new file mode 100644
index 0000000..79c8739
--- /dev/null
+++ b/src/vector_search_mcp/engine/qdrant_engine.py
@@ -0,0 +1,79 @@
+from collections.abc import Sequence
+from typing import final, override
+
+from qdrant_client import AsyncQdrantClient, models
+
+from ..config import Settings
+from ..models import SearchRow, Condition, Match, MatchAny, MatchExclude
+from .base_engine import BaseEngine
+
+__all__ = ["QdrantEngine"]
+
+
+@final
+class QdrantEngine(BaseEngine[list[models.ScoredPoint], models.Filter]):
+    def __init__(self) -> None:
+        self.settings = Settings()  # type: ignore[reportCallArgs]
+        self.client = AsyncQdrantClient(
+            url=self.settings.url, api_key=self.settings.api_key
+        )
+
+    @override
+    def transform_conditions(
+        self, conditions: list[Condition] | None
+    ) -> models.Filter | None:
+        if not conditions:
+            return None
+
+        filters: list[models.Condition] = []
+
+        for condition in conditions:
+            if isinstance(condition, Match):
+                filters.append(
+                    models.FieldCondition(
+                        key=condition.key,
+                        match=models.MatchValue(value=condition.value),
+                    )
+                )
+            elif isinstance(condition, MatchAny):
+                filters.append(
+                    models.FieldCondition(
+                        key=condition.key, match=models.MatchAny(any=condition.any)
+                    )
+                )
+            elif isinstance(condition, MatchExclude):
+                filters.append(
+                    models.FieldCondition(
+                        key=condition.key,
+                        match=models.MatchExcept(**{"except": condition.exclude}),
+                    )
+                )
+
+        return models.Filter(must=filters)
+
+    @override
+    def transform_response(self, response: list[models.ScoredPoint]) -> list[SearchRow]:
+        return [
+            SearchRow(chunk_id=str(point.id), score=point.score, payload=point.payload)
+            for point in response
+            if point.payload is not None
+        ]
+
+    @override
+    async def run_similarity_query(
+        self,
+        embedding: Sequence[float] | models.NamedVector,
+        collection: str,
+        limit: int = 10,
+        conditions: models.Filter | None = None,
+        threshold: float | None = None,
+    ) -> list[models.ScoredPoint]:
+        return await self.client.search(
+            collection_name=collection,
+            query_vector=embedding,
+            query_filter=conditions,
+            limit=limit,
+            with_payload=True,
+            with_vectors=False,
+            score_threshold=threshold,
+        )
diff --git a/src/vector_search_mcp/models.py b/src/vector_search_mcp/models.py
index cf6b84c..4727ddc 100644
--- a/src/vector_search_mcp/models.py
+++ b/src/vector_search_mcp/models.py
@@ -6,4 +6,22 @@ from pydantic import BaseModel
 class SearchRow(BaseModel):
     chunk_id: str
     score: float
-    payload: dict[str, Any]
+    payload: dict[str, Any]  # type: ignore[reportExplicitAny]
+
+
+class Condition(BaseModel): ...
+
+
+class Match(Condition):
+    key: str
+    value: str
+
+
+class MatchAny(Condition):
+    key: str
+    any: list[str]
+
+
+class MatchExclude(Condition):
+    key: str
+    exclude: list[str]