RAG mit LangChain: Python-Guide für den deutschen Mittelstand

# requirements.txt
# langchain, langchain-community, qdrant-client,
# sentence-transformers, pypdf, python-dotenv

from langchain_community.document_loaders import PyPDFLoader

loader = PyPDFLoader("technische_dokumentation.pdf")
docs = loader.load()
# ~50 pages → 200 chunks (500 tokens each)

from langchain_community.embeddings import OllamaEmbeddings

embeddings = OllamaEmbeddings(model="nomic-embed-text")
vectors = embeddings.embed_documents(chunk_texts)

from qdrant_client import QdrantClient

client = QdrantClient(path="./qdrant_db")  # Lokal!
client.create_collection(
    collection_name="firma_wissen",
    vectors_config={"size": 768}
)
client.upsert(collection_name="firma_wissen", documents=chunks)

from langchain_community.llms import Ollama
from langchain.chains import RetrievalQA

llm = Ollama(model="llama3.3:70b", temperature=0)
qa_chain = RetrievalQA.from_chain_type(
    llm=llm,
    chain_type="stuff",
    retriever=client.as_retriever()
)

# Anfrage
result = qa_chain.invoke(
    "Was sagt die technische Dokumentation über Wartungsintervalle?"
)
print(result["result"])