How to add reranking to RAG pipeline

pip install openai langchain

import os
from openai import OpenAI
from langchain_community.vectorstores import FAISS
from langchain_openai import ChatOpenAI, OpenAIEmbeddings
from langchain_core.prompts import ChatPromptTemplate

# Initialize OpenAI client
client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

# Sample documents and queries
documents = [
    {"id": "doc1", "text": "Python is a programming language."},
    {"id": "doc2", "text": "JavaScript is used for web development."},
    {"id": "doc3", "text": "RAG pipelines combine retrieval and generation."}
]

# Create embeddings for documents (simulate with dummy vectors here)
# In practice, use client.embeddings.create with a suitable model
embeddings = {
    "doc1": [0.1, 0.2, 0.3],
    "doc2": [0.4, 0.1, 0.5],
    "doc3": [0.3, 0.7, 0.2]
}

# Build a simple FAISS index (mock example, replace with real embeddings)
index = FAISS.from_texts([d["text"] for d in documents], OpenAIEmbeddings())

# Query
query = "What is Python?"

# Step 1: Retrieve top 3 documents
retrieved_docs = index.similarity_search(query, k=3)

# Step 2: Rerank retrieved docs using a reranker model
reranker_prompt_template = ChatPromptTemplate.from_template(
    "Given the query: {query}\nRank the following documents by relevance:\n{docs}\nRespond with the document texts sorted from most to least relevant."
)
reranker_prompt = reranker_prompt_template.format(query=query, docs="\n".join([doc.page_content for doc in retrieved_docs]))

reranker_response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": reranker_prompt}]
)

reranked_text = reranker_response.choices[0].message.content

# Step 3: Use top reranked document to generate answer
top_doc = reranked_text.split('\n')[0]  # Simplified extraction

generator_prompt = f"Answer the question: {query} using this document: {top_doc}"

generator_response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": generator_prompt}]
)

answer = generator_response.choices[0].message.content
print("Answer:", answer)