Last Updated: June 17, 2026
Level: Beginner** Time to complete**: 15 minutes** Goal**: After completing this tutorial, you’ll have learned how to build an indexing pipeline that will preprocess files based on their file type, using theFileTypeRouter
.
💡 (Optional): After creating the indexing pipeline in this tutorial, there is an optional section that shows you how to create a RAG pipeline on top of the document store you just created. You must have a
[Hugging Face API Key]for this section
Components Used #
: This component will help you route files based on their corresponding MIME type to different componentsFileTypeRouter
: This component will help you convert markdown files into Haystack DocumentsMarkdownToDocument
: This component will help you convert pdf files into Haystack DocumentsPyPDFToDocument
: This component will help you convert text files into Haystack DocumentsTextFileToDocument
(optional): This component will help you to make Documents more readable by removing extra whitespaces etc.DocumentCleaner
: This component will help you to split your Document into chunksDocumentSplitter
: This component will help you create embeddings for Documents.SentenceTransformersDocumentEmbedder
: This component will help you write Documents into the DocumentStoreDocumentWriter
Overview #
In this tutorial, you’ll build an indexing pipeline that preprocesses different types of files (markdown, txt and pdf). Each file will have its own FileConverter
. The rest of the indexing pipeline is fairly standard - split the documents into chunks, trim whitespace, create embeddings and write them to a Document Store.
Optionally, you can keep going to see how to use these documents in a query pipeline as well.
Installing dependencies #
%%bash
pip install haystack-ai huggingface-api-haystack
pip install sentence-transformers-haystack
pip install markdown-it-py mdit_plain pypdf
pip install gdown
Download All Files #
Files that you will use in this tutorial are stored in a GDrive folder. Either download files directly from the GDrive folder or run the code below. If you’re running this tutorial on colab, you’ll find the downloaded files under “/recipe_files” folder in “files” tab on the left.
Just like most real life data, these files are a mishmash of different types.
import gdown
url = "https://drive.google.com/drive/folders/1n9yqq5Gl_HWfND5bTlrCwAOycMDt5EMj"
output_dir = "recipe_files"
gdown.download_folder(url, quiet=True, output=output_dir)
Create a Pipeline to Index Documents #
Next, you’ll create a pipeline to index documents. To keep things uncomplicated, you’ll use an InMemoryDocumentStore
but this approach would also work with any other flavor of DocumentStore
.
You’ll need a different file converter class for each file type in our data sources: .pdf
, .txt
, and .md
in this case. Our FileTypeRouter
connects each file type to the proper converter.
from haystack.components.writers import DocumentWriter
from haystack.components.converters import MarkdownToDocument, PyPDFToDocument, TextFileToDocument
from haystack.components.preprocessors import DocumentSplitter, DocumentCleaner
from haystack.components.routers import FileTypeRouter
from haystack_integrations.components.embedders.sentence_transformers import SentenceTransformersDocumentEmbedder
from haystack import Pipeline
from haystack.document_stores.in_memory import InMemoryDocumentStore
document_store = InMemoryDocumentStore()
file_type_router = FileTypeRouter(mime_types=["text/plain", "application/pdf", "text/markdown"])
text_file_converter = TextFileToDocument()
markdown_converter = MarkdownToDocument()
pdf_converter = PyPDFToDocument()
From there, the steps to this indexing pipeline are a bit more standard. The DocumentCleaner
removes whitespace. Then this DocumentSplitter
breaks them into chunks of 150 words, with a bit of overlap to avoid missing context.
document_cleaner = DocumentCleaner()
document_splitter = DocumentSplitter(split_by="word", split_length=150, split_overlap=50)
Now you’ll add a SentenceTransformersDocumentEmbedder
to create embeddings from the documents. As the last step in this pipeline, the DocumentWriter
will write them to the InMemoryDocumentStore
.
document_embedder = SentenceTransformersDocumentEmbedder(model="sentence-transformers/all-MiniLM-L6-v2")
document_writer = DocumentWriter(document_store)
After creating all the components, add them to the indexing pipeline.
preprocessing_pipeline = Pipeline()
preprocessing_pipeline.add_component(instance=file_type_router, name="file_type_router")
preprocessing_pipeline.add_component(instance=text_file_converter, name="text_file_converter")
preprocessing_pipeline.add_component(instance=markdown_converter, name="markdown_converter")
preprocessing_pipeline.add_component(instance=pdf_converter, name="pypdf_converter")
preprocessing_pipeline.add_component(instance=document_cleaner, name="document_cleaner")
preprocessing_pipeline.add_component(instance=document_splitter, name="document_splitter")
preprocessing_pipeline.add_component(instance=document_embedder, name="document_embedder")
preprocessing_pipeline.add_component(instance=document_writer, name="document_writer")
Next, connect them 👇
preprocessing_pipeline.connect("file_type_router.text/plain", "text_file_converter.sources")
preprocessing_pipeline.connect("file_type_router.application/pdf", "pypdf_converter.sources")
preprocessing_pipeline.connect("file_type_router.text/markdown", "markdown_converter.sources")
preprocessing_pipeline.connect("text_file_converter", "document_cleaner")
preprocessing_pipeline.connect("pypdf_converter", "document_cleaner")
preprocessing_pipeline.connect("markdown_converter", "document_cleaner")
preprocessing_pipeline.connect("document_cleaner", "document_splitter")
preprocessing_pipeline.connect("document_splitter", "document_embedder")
preprocessing_pipeline.connect("document_embedder", "document_writer")
Let’s test this pipeline with a few recipes I’ve written. Are you getting hungry yet?
from pathlib import Path
preprocessing_pipeline.run({"file_type_router": {"sources": list(Path(output_dir).glob("**/*"))}})
🎉 If you only wanted to learn how to preprocess documents, you can stop here! If you want to see an example of using those documents in a RAG pipeline, read on.
(Optional) Build a pipeline to query documents #
Now, let’s build a RAG pipeline that answers queries based on the documents you just created in the section above. For this step, we will be using the HuggingFaceAPIChatGenerator so must have a
Hugging Face API Keyfor this section. We will be using the
Qwen/Qwen2.5-7B-Instruct
model.
import os
from getpass import getpass
if "HF_API_TOKEN" not in os.environ:
os.environ["HF_API_TOKEN"] = getpass("Enter Hugging Face token:")
In this step you’ll build a query pipeline to answer questions about the documents.
This pipeline takes the prompt, searches the document store for relevant documents, and passes those documents along to the LLM to formulate an answer.
⚠️ Notice how we used
sentence-transformers/all-MiniLM-L6-v2
to create embeddings for our documents before. This is why we will be using the same model to embed incoming questions.
from haystack_integrations.components.embedders.sentence_transformers import SentenceTransformersTextEmbedder
from haystack.components.retrievers.in_memory import InMemoryEmbeddingRetriever
from haystack.components.builders import ChatPromptBuilder
from haystack.dataclasses import ChatMessage
from haystack_integrations.components.generators.huggingface_api import HuggingFaceAPIChatGenerator
template = [
ChatMessage.from_user(
"""
Answer the questions based on the given context.
Context:
{% for document in documents %}
{{ document.content }}
{% endfor %}
Question: {{ question }}
Answer:
"""
)
]
pipe = Pipeline()
pipe.add_component("embedder", SentenceTransformersTextEmbedder(model="sentence-transformers/all-MiniLM-L6-v2"))
pipe.add_component("retriever", InMemoryEmbeddingRetriever(document_store=document_store))
pipe.add_component("chat_prompt_builder", ChatPromptBuilder(template=template))
pipe.add_component(
"llm",
HuggingFaceAPIChatGenerator(
api_type="serverless_inference_api",
api_params={"model": "Qwen/Qwen2.5-7B-Instruct", "provider": "together"}
),
)
pipe.connect("embedder.embedding", "retriever.query_embedding")
pipe.connect("retriever", "chat_prompt_builder.documents")
pipe.connect("chat_prompt_builder.prompt", "llm.messages")
Try it out yourself by running the code below. If all has gone well, you should have a complete shopping list from all the recipe sources. 🧂🥥🧄
question = (
"What ingredients would I need to make vegan keto eggplant lasagna, vegan persimmon flan, and vegan hemp cheese?"
)
pipe.run({"embedder": {"text": question}, "chat_prompt_builder": {"question": question}})
{’llm’: {‘replies’: [ChatMessage(_role=<ChatRole.ASSISTANT: ‘assistant’>, _content=[TextContent(text=‘To make vegan keto eggplant lasagna, vegan persimmon flan, and vegan hemp cheese, you would need the following ingredients:\n\n### Vegan Keto Eggplant Lasagna\n- 2 large eggplants\n- Salt (Hella salt)\n- 1/2 cup store-bought vegan mozzarella for topping\n- Pesto (ingredients: 4 oz basil, 1/4 cup almonds, 1/4 cup nutritional yeast, 1/4 cup olive oil, 1 recipe vegan pesto)\n- Spinach Tofu Ricotta (14 oz firm or extra firm tofu, 10 oz spinach, juice of 1 lemon, garlic powder to taste, salt to taste)\n- Macadamia Nut Cheese (1 cup macadamia nuts, 1/4 cup nutritional yeast, 1/4 cup olive oil, 1 recipe vegan pesto, 1 recipe spinach tofu ricotta, 1 tsp garlic powder, juice of half a lemon, salt to taste)\n\n### Vegan Persimmon Flan\n- 2 average-sized fuyu persimmons, strained\n- 1 tbsp cornstarch\n- 1/2 tsp agar agar\n- 1 tbsp agave nectar, or to taste\n- 2 tbsp granulated sugar\n- 1/4 cup coconut creme\n- 1/2 cup almond milk\n- 1/2 tsp vanilla\n\n### Vegan Hemp Cheese\n- 1/2 cup sunflower seeds\n- 1/2 cup hemp hearts\n- 1.5 teaspoons miso paste\n- 1 tsp nutritional yeast\n- 1/4 cup rejuvelac\n- 1/4th teaspoon salt, or to taste\n\n### Additional Tools and Notes\n- Casserole dish (9 x 13)\n- 2 ramekins\n- Blender or food processor\n- Saucepan\n- Immersion blender (optional)\n- Clean glass bowl\n- Rubber band\n- Dish towel\n- Knife\n- Hot water bath method (for flan)\n\nThese ingredients and tools will allow you to prepare all three dishes as described in the provided context.’)], _name=None, _meta={‘model’: ‘Qwen/Qwen2.5-7B-Instruct’, ‘finish_reason’: ‘stop’, ‘index’: 0, ‘usage’: {‘prompt_tokens’: 2031, ‘completion_tokens’: 446}})]}}
What’s next #
Congratulations on building an indexing pipeline that can preprocess different file types. Go forth and ingest all the messy real-world data into your workflows. 💥
If you liked this tutorial, you may also enjoy:
To stay up to date on the latest Haystack developments, you can sign up for our newsletter. Thanks for reading!