diff --git a/lucidia_llm/data/build_sft.py b/lucidia_llm/data/build_sft.py
new file mode 100644
index 0000000..ecff008
--- /dev/null
+++ b/lucidia_llm/data/build_sft.py
@@ -0,0 +1,6 @@
+"""Data builder for supervised fine-tuning dataset."""
+
+
+def build_sft_dataset(input_paths, output_path):
+    """Construct the SFT dataset from raw inputs (placeholder)."""
+    raise NotImplementedError("SFT dataset builder not implemented")