LangChain is a Python framework for applications that need more than a single language-model call. It connects components into chains: steps that can prepare a prompt, call a model, process its reply, or use another tool. For a simple task, a provider’s library may be enough; LangChain becomes more useful when the workflow has several connected parts. section 10.1 section 10.7
Classify customer reviews
Suppose a CSV file has a column of review text. A classification chain fills a prompt template with each review, asks a chat model whether its sentiment is positive or negative, and parses the reply into a string. Applying the chain to the reviews produces a result CSV with an added class column. The template specifies both the question and the permitted answers, rather than leaving the output format implicit. section 10.2.1 section 10.2.3
Listing 10.1 Sentiment classification using LangChain
from langchain_openai import ChatOpenAI
from langchain_core.prompts.chat import ChatPromptTemplate
from langchain_core.output_parsers.string import StrOutputParser
from langchain_core.runnables.passthrough import RunnablePassthrough
import argparse
import pandas as pd
def create_chain(): #1
""" Creates chain for text classification.
Returns:
a chain for text classification.
"""
prompt = ChatPromptTemplate.from_template( #2
'{text}\n'
'Is the sentiment positive or negative?\n'
'Answer ("Positive"/"Negative")\n')
llm = ChatOpenAI( #3
model='gpt-4o', temperature=0,
max_tokens=1)
parser = StrOutputParser() #4
#5
chain = ({'text':RunnablePassthrough()} | prompt | llm | parser)
return chain
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('file_path', type=str, help='Path to input file')
args = parser.parse_args()
df = pd.read_csv(args.file_path) #6
chain = create_chain() #7
results = chain.batch(list(df['text'])) #8
df['class'] = results #9
df.to_csv('result.csv')
The Python listing shows how a prompt template, ChatOpenAI model, and string output parser form the review-classification chain. section 10.2.3
When the next step depends on the question, an agent offers a different approach. Given suitable tools, it can choose which data source to access and in what order—for example, query an SQLite sales database and then search the web using a title found in the results. LangChain provides constructors for SQL agents and a way to load a web-search tool; that search tool also requires a SerpAPI account and access key. section 10.4 section 10.4.2
To try the classification example, install compatible LangChain and provider packages, then run the script against a reviews CSV and inspect its output file. Pinning versions matters here: LangChain’s interfaces were changing rapidly when these examples were written. section 10.1 section 10.2.4
