diff --git a/pkgs/classgen/main.py b/pkgs/classgen/main.py
index 76bb115f0..856b1c545 100644
--- a/pkgs/classgen/main.py
+++ b/pkgs/classgen/main.py
@@ -65,8 +65,8 @@ def field_def_from_default_type(
     field_name: str,
     field_types: set[str],
     class_name: str,
-    finalize_field: Callable[..., str],
-) -> str | None:
+    finalize_field: Callable[..., tuple[str, str]],
+) -> tuple[str, str] | None:
     if "dict" in str(field_types):
         return finalize_field(
             field_types=field_types,
@@ -127,8 +127,8 @@ def field_def_from_default_value(
     field_name: str,
     field_types: set[str],
     nested_class_name: str,
-    finalize_field: Callable[..., str],
-) -> str | None:
+    finalize_field: Callable[..., tuple[str, str]],
+) -> tuple[str, str] | None:
     # default_value = prop_info.get("default")
     if default_value is None:
         return finalize_field(
@@ -184,7 +184,7 @@ def get_field_def(
     default: str | None = None,
     default_factory: str | None = None,
     type_appendix: str = "",
-) -> str:
+) -> tuple[str, str]:
     if "None" in field_types or default or default_factory:
         if "None" in field_types:
             field_types.remove("None")
@@ -193,7 +193,7 @@ def get_field_def(
     else:
         serialised_types = " | ".join(field_types) + type_appendix
 
-    return f"{field_name}: {serialised_types}"
+    return (field_name, serialised_types)
 
 
 # Recursive function to generate dataclasses from JSON schema
@@ -204,8 +204,8 @@ def generate_dataclass(
 ) -> str:
     properties = schema.get("properties", {})
 
-    required_fields = []
-    fields_with_default = []
+    required_fields: list[tuple[str, str]] = []
+    fields_with_default: list[tuple[str, str]] = []
     nested_classes: list[str] = []
 
     # if We are at the top level, and the attribute name is in shallow
@@ -218,7 +218,7 @@ def generate_dataclass(
         field_name = prop.replace("-", "_")
 
         if len(attr_path) == 0 and prop not in attrs:
-            field_def = f"{field_name}: NotRequired[dict[str, Any]]"
+            field_def = field_name, "NotRequired[dict[str, Any]]"
             fields_with_default.append(field_def)
             # breakpoint()
             continue
@@ -345,9 +345,18 @@ def generate_dataclass(
             )
             required_fields.append(field_def)
 
-    # breakpoint()
-
-    fields_str = "\n    ".join(required_fields + fields_with_default)
+    # Join field name with type to form a complete field declaration
+    # e.g. "name: str"
+    all_field_declarations = [
+        f"{n}: {t}" for n, t in (required_fields + fields_with_default)
+    ]
+    hoisted_types: str = "\n".join(
+        [
+            f"{class_name}{n.capitalize()}Type = {x}"
+            for n, x in (required_fields + fields_with_default)
+        ]
+    )
+    fields_str = "\n    ".join(all_field_declarations)
     nested_classes_str = "\n\n".join(nested_classes)
 
     class_def = f"\nclass {class_name}(TypedDict):\n"
@@ -356,6 +365,8 @@ def generate_dataclass(
     else:
         class_def += f"    {fields_str}"
 
+    class_def += f"\n\n{hoisted_types}\n"
+
     return f"{nested_classes_str}\n\n{class_def}" if nested_classes_str else class_def