Skip to main content

​
Overview

The chains.py module sets up the complete RAG (Retrieval-Augmented Generation) pipeline, including the vector store, retriever, LLM, and chain components.

​
Module Exports

The module exports the following variables that can be imported by other modules:
  • rag_chain - Complete RAG pipeline
  • retriever - MMR-based document retriever
  • llm - ChatOpenAI instance
  • db - Chroma vector store
  • embeddings - HuggingFace embeddings model

​
Variables

​
db

​
db
Chroma
The Chroma vector store loaded from the persisted database. Contains all Hadith document embeddings and metadata.
db, embeddings = load_and_prepare_data()

​
embeddings

​
embeddings
HuggingFaceEmbeddings
HuggingFace embeddings model using sentence-transformers/all-MiniLM-L6-v2 for vectorization.

​
retriever

​
retriever
VectorStoreRetriever
MMR (Maximal Marginal Relevance) based retriever that fetches diverse, relevant documents from the vector store.

​
Configuration

​
search_type
string
default:"mmr"
Search algorithm type. Uses Maximal Marginal Relevance to balance relevance and diversity.
​
k
integer
default:"4"
Number of documents to return in retrieval results.
​
fetch_k
integer
default:"10"
Number of candidate documents to fetch before MMR re-ranking.
retriever = db.as_retriever(
    search_type="mmr",
    search_kwargs={"k": 4, "fetch_k": 10}
)

​
llm

​
llm
ChatOpenAI
Language model instance using DeepSeek Chat v3 through OpenRouter API.

​
Configuration

​
model
string
default:"deepseek/deepseek-chat-v3-0324:free"
Model identifier for DeepSeek Chat v3 (free tier).
​
base_url
string
default:"https://openrouter.ai/api/v1"
OpenRouter API endpoint for model access.
llm = ChatOpenAI(
    model="deepseek/deepseek-chat-v3-0324:free",
    base_url="https://openrouter.ai/api/v1"
)
Requires OPENAI_API_KEY environment variable to be set with your OpenRouter API key.

​
rag_chain

​
rag_chain
RunnableSequence
Complete RAG pipeline that retrieves relevant documents and generates answers using the LLM.
The RAG chain is composed of:
  1. Retriever: Fetches relevant Hadith documents
  2. Document Chain: Combines retrieved documents with the prompt template
  3. LLM: Generates the final answer
question_answer_chain = create_stuff_documents_chain(llm, qa_prompt)
rag_chain = create_retrieval_chain(retriever, question_answer_chain)

​
Usage Example

from chains import rag_chain

# Invoke the RAG chain with a question
response = rag_chain.invoke({
    "input": "What does Islam say about charity?",
    "chat_history": []
})

# Access the generated answer
print(response["answer"])

# Access retrieved context documents
print(response["context"])

​
Input Format

​
input
string
required
The user’s question or query.
​
chat_history
list
required
List of previous messages for conversation context. Each message should be a dictionary with role and content keys.
{
    "input": "What are the pillars of Islam?",
    "chat_history": [
        {"role": "user", "content": "Previous question"},
        {"role": "assistant", "content": "Previous answer"}
    ]
}

​
Output Format

​
answer
string
The generated response from the LLM based on retrieved Hadith documents.
​
context
list[Document]
List of retrieved documents used to generate the answer.
​
input
string
Echo of the original input question.

​
Dependencies

  • langchain.chains.create_retrieval_chain
  • langchain.chains.combine_documents.create_stuff_documents_chain
  • langchain_openai.ChatOpenAI
  • dotenv.load_dotenv
  • loader.load_and_prepare_data
  • prompts.qa_prompt

​
Environment Variables

​
OPENAI_API_KEY
string
required
Your OpenRouter API key for accessing the DeepSeek model.
Create a .env file in your project root: 
OPENAI_API_KEY=your_openrouter_api_key_here

Build docs developers (and LLMs) love

Get started for freeTalk to us