diff --git a/.github/workflows/tests.yml b/.github/workflows/tests.yml
index 303f1bc..5e2f0b1 100644
--- a/.github/workflows/tests.yml
+++ b/.github/workflows/tests.yml
@@ -40,10 +40,15 @@ jobs:
       run: pip install pytest pydantic numpy
     
     - name: Run Python tests
-      run: pytest tests/ -v --tb=short
+      run: |
+        # Clear PYTHONPATH to ensure we test the installed wheel, not the local source
+        export PYTHONPATH=
+        pytest tests/ -v --tb=short
     
     - name: Run benchmarks (smoke test)
-      run: python benchmarks/test_10k.py
+      run: |
+        export PYTHONPATH=
+        python benchmarks/test_10k.py
 
   test-typescript:
     name: TypeScript Tests
diff --git a/pyproject.toml b/pyproject.toml
index 398679b..534b367 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -76,4 +76,4 @@ known-first-party = ["b_fast"]
 
 [tool.maturin]
 python-source = "python"
-module-name = "b_fast.b_fast"
+module-name = "b_fast._b_fast"
diff --git a/python/b_fast/__init__.py b/python/b_fast/__init__.py
index db323c6..a61c2bc 100644
--- a/python/b_fast/__init__.py
+++ b/python/b_fast/__init__.py
@@ -4,7 +4,7 @@
 Ultra-fast binary serialization library with Rust backend.
 """
 
-from .b_fast import BFast
+from ._b_fast import BFast
 from .integration import BFastResponse
 
 __version__ = "1.2.0"
diff --git a/python/b_fast/_b_fast.abi3.so b/python/b_fast/_b_fast.abi3.so
new file mode 100755
index 0000000..91b8a1b
Binary files /dev/null and b/python/b_fast/_b_fast.abi3.so differ
diff --git a/python/b_fast/b_fast.abi3.so b/python/b_fast/b_fast.abi3.so
new file mode 100755
index 0000000..998977d
Binary files /dev/null and b/python/b_fast/b_fast.abi3.so differ
diff --git a/python/b_fast/integration.py b/python/b_fast/integration.py
index 567a4ab..5d19cfb 100644
--- a/python/b_fast/integration.py
+++ b/python/b_fast/integration.py
@@ -1,16 +1,28 @@
 from typing import Any
 
-from fastapi import Response
-
-import b_fast
+try:
+    from fastapi import Response
+    FASTAPI_AVAILABLE = True
+except ImportError:
+    # Fallback class to prevent crash on import
+    class Response: 
+        def __init__(self, *args, **kwargs):
+            pass
+    FASTAPI_AVAILABLE = False
 
 
 class BFastResponse(Response):
     media_type = "application/x-bfast"
 
     def __init__(self, content: Any, **kwargs):
+        if not FASTAPI_AVAILABLE:
+            raise ImportError("FastAPI is required to use BFastResponse. Install it with 'pip install fastapi'.")
+        
+        # Delayed import to avoid circular dependency
+        from ._b_fast import BFast
+        
         super().__init__(content=content, **kwargs)
-        self.encoder = b_fast.BFast()
+        self.encoder = BFast()
 
     def render(self, content: Any) -> bytes:
         return self.encoder.encode_packed(content, compress=True)
diff --git a/src/lib.rs b/src/lib.rs
index 364448c..53d35ca 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -27,7 +27,6 @@ const TAG_UUID: u8 = 0xD4;
 const TAG_DECIMAL: u8 = 0xD5;
 
 #[allow(non_local_definitions)]
-#[repr(align(64))]
 #[pyclass]
 pub struct BFast {
     string_table: AHashMap<String, u32>,
@@ -796,7 +795,7 @@ impl BFast {
 }
 
 #[pymodule]
-fn b_fast(_py: Python, m: &PyModule) -> PyResult<()> {
+fn _b_fast(_py: Python, m: &PyModule) -> PyResult<()> {
     m.add_class::<BFast>()?;
     m.add(
         "BFastError",