[POC] Expose retrieval SDG as a Data Designer recipe

plugins/data-designer-retrieval-sdg/README.md

@@ -64,6 +64,24 @@ data-designer-retrieval-sdg generate \
     --num-pairs 7
 ```
 
+### Preview through Data Designer recipes
+
+When installed with a Data Designer CLI that supports recipe entry
+points, this package also registers `retrieval-sdg` under
+`data_designer.recipes`:
+
+```bash
+data-designer preview --recipe retrieval-sdg -- \
+    --input-dir ./my_documents \
+    --num-pairs 2
+```
+
+Recipe-specific options can be inspected without running generation:
+
+```bash
+data-designer recipes help retrieval-sdg
+```
+
 ### Convert to training format
 
 ```bash

plugins/data-designer-retrieval-sdg/pyproject.toml

@@ -27,6 +27,9 @@ classifiers = [
 embedding-dedup = "data_designer_retrieval_sdg.plugins:embedding_dedup_plugin"
 document-chunker = "data_designer_retrieval_sdg.plugins:document_chunker_plugin"
 
+[project.entry-points."data_designer.recipes"]
+retrieval-sdg = "data_designer_retrieval_sdg.recipe:load_config_builder"
+
 [project.scripts]
 data-designer-retrieval-sdg = "data_designer_retrieval_sdg.cli:main"
 

plugins/data-designer-retrieval-sdg/src/data_designer_retrieval_sdg/config.py

@@ -54,6 +54,7 @@ def side_effect_columns(self) -> list[str]:
         """Additional columns produced as side effects."""
         return []
 
-    def get_column_emoji(self) -> str:
+    @staticmethod
+    def get_column_emoji() -> str:
         """Emoji displayed in logs for this column type."""
         return "🔍"

plugins/data-designer-retrieval-sdg/src/data_designer_retrieval_sdg/recipe.py

@@ -0,0 +1,170 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""Typer-backed Data Designer recipe entry point for retrieval SDG."""
+
+from __future__ import annotations
+
+from pathlib import Path
+from typing import Annotated
+
+import click
+import data_designer.config as dd
+import typer
+
+from data_designer_retrieval_sdg.pipeline import (
+    DEFAULT_CHAT_MODEL,
+    DEFAULT_EMBED_MODEL,
+    DEFAULT_PROVIDER,
+    build_qa_generation_pipeline,
+)
+from data_designer_retrieval_sdg.seed_source import DocumentChunkerSeedSource
+
+
+def load_config_builder(params: dd.DataDesignerScriptParams | None = None) -> dd.DataDesignerConfigBuilder:
+    """Build the retrieval SDG pipeline from forwarded Data Designer CLI args.
+
+    Args:
+        params: Data Designer script parameters. ``params.argv`` contains the
+            arguments supplied after ``data-designer preview/create --recipe
+            retrieval-sdg --``.
+
+    Returns:
+        A configured Data Designer config builder for retrieval SDG generation.
+    """
+    argv = list(tuple(getattr(params, "argv", ())))
+    command = typer.main.get_command(build_typer_app())
+    config_builder = command.main(
+        args=argv,
+        prog_name="data-designer preview/create --recipe retrieval-sdg --",
+        standalone_mode=False,
+    )
+
+    if config_builder == 0 and any(arg in {"--help", "-h"} for arg in argv):
+        raise SystemExit(0)
+    if not isinstance(config_builder, dd.DataDesignerConfigBuilder):
+        raise TypeError(f"Recipe returned {type(config_builder).__name__}, expected DataDesignerConfigBuilder")
+    return config_builder
+
+
+def build_typer_app() -> typer.Typer:
+    """Build the Typer app used for recipe inspection and execution.
+
+    Returns:
+        Typer app describing the retrieval SDG recipe interface.
+    """
+    app = typer.Typer(add_completion=False, help="Build the retrieval SDG Data Designer workflow.")
+    app.command(name=None, help="Build the retrieval SDG Data Designer workflow.")(recipe_command)
+    return app
+
+
+def recipe_command(
+    input_dir: Annotated[Path, typer.Option("--input-dir", help="Directory containing text files")],
+    file_pattern: Annotated[str, typer.Option("--file-pattern", help="Filename glob (basenames only)")] = "*",
+    recursive: Annotated[
+        bool,
+        typer.Option("--recursive/--no-recursive", help="Enable recursive search"),
+    ] = True,
+    file_extensions: Annotated[
+        list[str] | None,
+        typer.Option(
+            "--file-extensions",
+            help="Allowed file extensions (use empty string '' to match files without extensions)",
+        ),
+    ] = None,
+    min_text_length: Annotated[int, typer.Option("--min-text-length", help="Minimum document text length")] = 50,
+    sentences_per_chunk: Annotated[int, typer.Option("--sentences-per-chunk", help="Sentences per chunk")] = 5,
+    num_sections: Annotated[int, typer.Option("--num-sections", help="Sections to divide chunks into")] = 1,
+    num_files: Annotated[int | None, typer.Option("--num-files", help="Max files to process")] = None,
+    multi_doc: Annotated[bool, typer.Option("--multi-doc", help="Enable multi-doc bundling")] = False,
+    bundle_size: Annotated[int, typer.Option("--bundle-size", help="Docs per bundle")] = 2,
+    bundle_strategy: Annotated[
+        str,
+        typer.Option(
+            "--bundle-strategy",
+            help="Section splitting strategy",
+            click_type=click.Choice(["sequential", "doc_balanced", "interleaved"]),
+        ),
+    ] = "sequential",
+    max_docs_per_bundle: Annotated[int, typer.Option("--max-docs-per-bundle", help="Max docs per bundle")] = 3,
+    multi_doc_manifest: Annotated[
+        Path | None, typer.Option("--multi-doc-manifest", help="Manifest for explicit bundles")
+    ] = None,
+    start_index: Annotated[int, typer.Option("--start-index", help="Start seed row index")] = 0,
+    end_index: Annotated[int, typer.Option("--end-index", help="End seed row index")] = 199,
+    max_artifacts_per_type: Annotated[int, typer.Option("--max-artifacts-per-type", help="Max artifacts per type")] = 2,
+    num_pairs: Annotated[int, typer.Option("--num-pairs", help="QA pairs per document")] = 7,
+    min_hops: Annotated[int, typer.Option("--min-hops", help="Min hops for multi-hop questions")] = 2,
+    max_hops: Annotated[int, typer.Option("--max-hops", help="Max hops for multi-hop questions")] = 4,
+    min_complexity: Annotated[int, typer.Option("--min-complexity", help="Min question complexity")] = 4,
+    similarity_threshold: Annotated[
+        float, typer.Option("--similarity-threshold", help="Cosine threshold for QA-pair dedup")
+    ] = 0.9,
+    artifact_extraction_model: Annotated[
+        str, typer.Option("--artifact-extraction-model", help="Artifact extraction model")
+    ] = DEFAULT_CHAT_MODEL,
+    artifact_extraction_provider: Annotated[
+        str, typer.Option("--artifact-extraction-provider", help="Artifact extraction provider")
+    ] = DEFAULT_PROVIDER,
+    qa_generation_model: Annotated[str, typer.Option("--qa-generation-model", help="QA generation model")] = (
+        DEFAULT_CHAT_MODEL
+    ),
+    qa_generation_provider: Annotated[str, typer.Option("--qa-generation-provider", help="QA generation provider")] = (
+        DEFAULT_PROVIDER
+    ),
+    quality_judge_model: Annotated[str, typer.Option("--quality-judge-model", help="Quality judge model")] = (
+        DEFAULT_CHAT_MODEL
+    ),
+    quality_judge_provider: Annotated[str, typer.Option("--quality-judge-provider", help="Quality judge provider")] = (
+        DEFAULT_PROVIDER
+    ),
+    embed_model: Annotated[str, typer.Option("--embed-model", help="Embedding model")] = DEFAULT_EMBED_MODEL,
+    embed_provider: Annotated[str, typer.Option("--embed-provider", help="Embedding provider")] = DEFAULT_PROVIDER,
+    max_parallel_requests_for_gen: Annotated[
+        int | None, typer.Option("--max-parallel-requests-for-gen", help="Max parallel generation requests")
+    ] = None,
+) -> dd.DataDesignerConfigBuilder:
+    """Build the retrieval SDG Data Designer workflow.
+
+    Returns:
+        A configured Data Designer config builder.
+    """
+    if end_index < start_index:
+        raise click.BadParameter("--end-index must be greater than or equal to --start-index")
+
+    seed_source = DocumentChunkerSeedSource(
+        path=str(input_dir),
+        file_pattern=file_pattern,
+        recursive=recursive,
+        file_extensions=file_extensions or [".txt", ".md", ".text"],
+        min_text_length=min_text_length,
+        sentences_per_chunk=sentences_per_chunk,
+        num_sections=num_sections,
+        num_files=num_files,
+        multi_doc=multi_doc,
+        bundle_size=bundle_size,
+        bundle_strategy=bundle_strategy,
+        max_docs_per_bundle=max_docs_per_bundle,
+        multi_doc_manifest=str(multi_doc_manifest) if multi_doc_manifest else None,
+    )
+
+    return build_qa_generation_pipeline(
+        seed_source=seed_source,
+        start_index=start_index,
+        end_index=end_index,
+        max_artifacts_per_type=max_artifacts_per_type,
+        num_pairs=num_pairs,
+        min_hops=min_hops,
+        max_hops=max_hops,
+        min_complexity=min_complexity,
+        similarity_threshold=similarity_threshold,
+        max_parallel_requests_for_gen=max_parallel_requests_for_gen,
+        artifact_extraction_model=artifact_extraction_model,
+        artifact_extraction_provider=artifact_extraction_provider,
+        qa_generation_model=qa_generation_model,
+        qa_generation_provider=qa_generation_provider,
+        quality_judge_model=quality_judge_model,
+        quality_judge_provider=quality_judge_provider,
+        embed_model=embed_model,
+        embed_provider=embed_provider,
+    )

plugins/data-designer-retrieval-sdg/tests/test_recipe.py

@@ -0,0 +1,61 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+from __future__ import annotations
+
+from pathlib import Path
+from types import SimpleNamespace
+
+import typer
+from click.testing import CliRunner
+from data_designer.config.config_builder import DataDesignerConfigBuilder
+
+from data_designer_retrieval_sdg.recipe import build_typer_app, load_config_builder
+from data_designer_retrieval_sdg.seed_source import DocumentChunkerSeedSource
+
+
+def test_load_config_builder_builds_retrieval_sdg_pipeline(tmp_path: Path) -> None:
+    """The recipe entry point builds the retrieval SDG workflow from argv."""
+    params = SimpleNamespace(
+        argv=(
+            "--input-dir",
+            str(tmp_path),
+            "--num-pairs",
+            "2",
+            "--start-index",
+            "1",
+            "--end-index",
+            "4",
+            "--file-extensions",
+            ".txt",
+        )
+    )
+
+    builder = load_config_builder(params)
+
+    assert isinstance(builder, DataDesignerConfigBuilder)
+    seed_config = builder.get_seed_config()
+    assert seed_config is not None
+    assert isinstance(seed_config.source, DocumentChunkerSeedSource)
+    assert seed_config.source.path == str(tmp_path)
+    assert seed_config.source.file_extensions == [".txt"]
+    assert seed_config.selection_strategy is not None
+    assert seed_config.selection_strategy.start == 1
+    assert seed_config.selection_strategy.end == 4
+    assert [column.name for column in builder.get_column_configs()] == [
+        "document_artifacts",
+        "qa_generation",
+        "deduplicated_qa_pairs",
+        "qa_evaluations",
+    ]
+
+
+def test_build_typer_app_exposes_recipe_help() -> None:
+    """The recipe exposes Typer metadata for Data Designer inspection."""
+    command = typer.main.get_command(build_typer_app())
+    result = CliRunner().invoke(command, ["--help"])
+
+    assert result.exit_code == 0
+    assert "Build the retrieval SDG Data Designer workflow." in result.output
+    assert "--input-dir" in result.output
+    assert "--num-pairs" in result.output