File size: 2,223 Bytes
7813359
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
35f4ec7
7813359
 
04b8e0f
7813359
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
from openai import OpenAI
from langchain_community.vectorstores import FAISS as LangChainFAISS
from langchain_openai import OpenAIEmbeddings
import streamlit as st
from functools import lru_cache

# Initialize OpenAI client with Streamlit secrets
client = OpenAI(api_key=st.secrets["OPENAI_API_KEY"])

# Cache the vector store loading
@lru_cache(maxsize=1)
def load_vector_store():
    embeddings = OpenAIEmbeddings(model="text-embedding-ada-002", openai_api_key=st.secrets["OPENAI_API_KEY"])
    return LangChainFAISS.load_local(
        folder_path="faiss_index",
        embeddings=embeddings,
        allow_dangerous_deserialization=True
    )

vector_store = load_vector_store()

def query_rag(query, top_k=8):
    results = vector_store.similarity_search(query, k=top_k)
    context = ""
    for i, doc in enumerate(results):
        meta = doc.metadata
        context += f"Source: {meta['source']}\n"
        if meta["part"]:
            context += f"Part: {meta['part']}\n"
        context += f"Heading: {meta['heading']}\n"
        if meta["title"]:
            context += f"Title: {meta['title']}\n"
        if meta["sub_title"]:
            context += f"Sub-title: {meta['sub_title']}\n"
        if meta["paragraph_number"]:
            context += f"Paragraph {meta['paragraph_number']}"
        if meta["paragraph_title"]:
            context += f": {meta['paragraph_title']}\n"
        elif meta["paragraph_number"]:
            context += "\n"
        if meta["sub_para_title"]:
            context += f"Sub-paragraph: {meta['sub_para_title']}\n"
        context += f"Text: {doc.page_content}\n\n"
    
    prompt = f"""
    User Query: {query}
    Retrieved Context:
    {context}
    
    Provide a clear, easy-to-understand explanation based on the context as you are explaining to a average person. Include direct quotes with citations (e.g., 'Part I, Paragraph 1: Article I. Declaration of Union') where relevant. Structure the response by grouping information from the same 'part' together.
    """
    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": prompt}],
        max_tokens=3000
    )
    return response.choices[0].message.content