A retriever is the RAG stage that finds information relevant to a query and supplies document chunks to downstream context assembly. Retrieval quality directly affects the answers an LLM can produce, so the component must be treated as a production integration rather than a standalone similarity lookup. section 12.2.4 section 12.2.5
A practical retriever pipeline typically:
- Loads source Markdown files recursively and splits them into overlapping chunks, preserving useful document structure through configured separators. section 12.2.4
- Embeds document chunks with a document-specific retrieval task and stores the vectors in a FAISS index. section 12.2.4
- Persists the index and reloads it with the same embedding model, because dimension, normalization, or model-version changes can break compatibility. section 12.2.5
- Exposes similarity search through a retriever interface, allowing a query to produce documents that can be joined into LLM context. section 12.2.5
Retrieving an encryption-key procedure
A query such as “How do I rotate encryption keys?” is passed to the retriever, which returns the most relevant indexed chunks. The chunks are joined into context before being supplied to the LLM. section 12.2.4 section 12.2.5
Listing 12.9 Persistent FAISS index and LangChain retriever
from langchain_community.vectorstores import FAISS
from index import GeminiEmbeddings
embeddings = GeminiEmbeddings() #1
vector_store = FAISS.load_local(
"DATA/faiss_index",
embeddings=embeddings,
allow_dangerous_deserialization=True #2
)
retriever = vector_store.as_retriever() #3
Shows how a persistent FAISS index becomes a LangChain retriever after being loaded with the matching embedding wrapper. section 12.2.5
Listing 12.10 RAG query processing with retrieval and context
query = "How do I rotate encryption keys?" #1
docs = retriever.invoke(query) #2
context = "\n\n".join( #3
[doc.page_content for doc in docs]) #3
Shows the runtime handoff from a user query to retrieved documents and assembled context. section 12.2.5
