fix(python): compose Postgres filters as SQL

python/semantic_kernel/connectors/postgres.py

@@ -794,11 +794,12 @@ def _construct_vector_query(
         )
 
         if where_clauses := self._build_filter(options.filter):  # type: ignore
-            query += (
-                sql.SQL("WHERE {clause}").format(clause=sql.SQL(" AND ").join(where_clauses))
+            clause = (
+                sql.SQL(" AND ").join(sql.SQL(where_clause) for where_clause in where_clauses)
                 if isinstance(where_clauses, list)
-                else sql.SQL("WHERE {clause}").format(clause=where_clauses)
+                else sql.SQL(where_clauses)
             )
+            query += sql.SQL(" WHERE {clause}").format(clause=clause)
 
         query += sql.SQL(" ORDER BY {dist_col} LIMIT {limit}").format(
             dist_col=sql.Identifier(self._distance_column_name),

python/tests/unit/connectors/memory/test_postgres_store.py

@@ -17,7 +17,13 @@
     PostgresSettings,
     PostgresStore,
 )
-from semantic_kernel.data.vector import DistanceFunction, IndexKind, VectorStoreField, vectorstoremodel
+from semantic_kernel.data.vector import (
+    DistanceFunction,
+    IndexKind,
+    VectorSearchOptions,
+    VectorStoreField,
+    vectorstoremodel,
+)
 
 
 @fixture(scope="function")
@@ -322,6 +328,32 @@ def model_post_init(self, context: Any) -> None:
     assert statement_str == expected_statement
 
 
+@pytest.mark.parametrize(
+    "filter_, expected_clause",
+    [
+        ("lambda x: x.id == 1", 'WHERE "id" = 1'),
+        (["lambda x: x.id == 1", "lambda x: x.id > 0"], 'WHERE "id" = 1 AND "id" > 0'),
+    ],
+)
+def test_vector_search_filter_is_composed_as_sql(filter_, expected_clause) -> None:
+    """Filter predicates must remain SQL fragments instead of quoted string literals."""
+    pool = AsyncConnectionPool(open=False)
+    collection = PostgresCollection(
+        collection_name="test_collection",
+        record_type=SimpleDataModel,
+        connection_pool=pool,
+    )
+
+    query, _, _ = collection._construct_vector_query(
+        vector=[1.0, 2.0, 3.0],
+        options=VectorSearchOptions(filter=filter_, top=3),
+    )
+
+    query_string = query.as_string()
+    assert f'FROM "public"."test_collection" {expected_clause} ORDER BY' in query_string
+    assert '\'"id"' not in query_string
+
+
 async def test_model_post_init_conflicting_distance_column_name(vector_store: PostgresStore) -> None:
     @vectorstoremodel
     @dataclass