A data-processing pipeline connects steps so that one step’s result becomes input to the next. It is useful when a task needs more than a single language-model call—for example, when data must be prepared, analyzed, and turned into a usable output. section 10.1 section 10.0
Classify customer feedback
A simple text-classification pipeline takes a comment, inserts it into a prompt asking for “Positive” or “Negative,” sends the prompt to GPT-4o, and parses the reply into an answer string. The same chain can process a batch of comments. In a complete Python script, comments come from a CSV file and the results are saved in a new class column. section 10.2.1 section 10.2.2 section 10.2.3
from langchain_core.runnables.passthrough import RunnablePassthrough
chain = ({'text':RunnablePassthrough()} | prompt | llm | parser)
The Python expression connects input mapping, prompt template, model, and parser in the order each result is needed. section 10.2.2
The same pattern supports different kinds of work:
- Speech translation records audio, transcribes it, translates the text into a requested language, then generates spoken output. section 7.4.1
- Question answering over large data loads and divides documents, indexes their chunks, retrieves relevant items, and includes them with the question in a model prompt. section 10.6.1
Start with a few clear steps and check what each one receives and returns. LangChain calls such connected components a chain; its chains can also support parallel or conditional steps, though neither is needed for the simple classifier. The illustrated setup uses specific package versions because LangChain interfaces may change. section 10.2.1 section 10.1
