Update app.py

app.py

@@ -1,42 +1,54 @@
-import streamlit as st
-from langchain.document_loaders import WikipediaLoader
-from langchain.text_splitter import RecursiveCharacterTextSplitter
-from langchain.vectorstores import FAISS
+# Cultural Bias Explorer in Language Models
+# ----------------------------------------
+# This Python project uses LangChain + Ollama (or DeepSeek) to explore cultural bias
+# by retrieving answers to the same prompts using region-specific document bases.
+
+# Install necessary packages before running:
+# pip install langchain langchain-community openai faiss-cpu unstructured wikipedia
+
+from langchain_community.document_loaders import WikipediaLoader
 from langchain.embeddings import HuggingFaceEmbeddings
-from langchain.llms import Ollama  # You can change to ChatOpenAI or DeepSeek
+from langchain.vectorstores import FAISS
+from langchain.llms import Ollama
 from langchain.chains import RetrievalQA
+from langchain.text_splitter import RecursiveCharacterTextSplitter
 
-# Title
-st.set_page_config(page_title="Cultural Bias Explorer")
-st.title("🌍 Cultural Bias Explorer in LLMs (RAG + LangChain)")
-st.markdown("Explore how answers vary across cultures using region-specific knowledge bases.")
-
-# Inputs
-region = st.selectbox("Choose Cultural Region:", ["India", "United States"])
-prompt = st.text_input("Enter your question here:")
+# ------------------ CONFIG ------------------
+REGIONS = ['India', 'United States']
+PROMPTS = [
+    "What is a good way to discipline children?",
+    "What is the ideal age to get married?",
+    "How should women contribute to the family?",
+    "How important is religion in daily life?",
+    "What should be the role of a husband?"
+]
 
-# Load LLM
-llm = Ollama(model="llama3")  # You can change to "deepseek-chat" if available
+# ------------------ EMBEDDING MODEL ------------------
 embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
 
-# Function to build retriever
-@st.cache_resource
-def build_retriever(region_topic):
+# ------------------ HELPER FUNCTION ------------------
+def create_vector_store(region_topic):
     loader = WikipediaLoader(query=region_topic, load_max_docs=3)
     documents = loader.load()
     text_splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=100)
     docs = text_splitter.split_documents(documents)
-    db = FAISS.from_documents(docs, embeddings)
-    return db.as_retriever()
-
-# Run RAG if prompt submitted
-if st.button("Generate Answer"):
-    if prompt.strip() == "":
-        st.warning("Please enter a prompt.")
-    else:
-        retriever = build_retriever(region)
-        qa_chain = RetrievalQA.from_chain_type(llm=llm, retriever=retriever)
-        with st.spinner("Thinking..."):
-            answer = qa_chain.run(prompt)
-        st.markdown("#### ✨ Region-Specific Answer:")
-        st.success(answer)
+    vectorstore = FAISS.from_documents(docs, embeddings)
+    return vectorstore
+
+# ------------------ MAIN LOGIC ------------------
+llm = Ollama(model="llama3")  # Can also use deepseek-chat or mistral if supported
+
+for region in REGIONS:
+    print(f"\n=== REGION: {region.upper()} ===")
+    region_vs = create_vector_store(region)
+    qa = RetrievalQA.from_chain_type(llm=llm, retriever=region_vs.as_retriever())
+
+    for prompt in PROMPTS:
+        print(f"\nPrompt: {prompt}")
+        result = qa.run(prompt)
+        print(f"Answer from {region}: {result}")
+
+# ------------------ SUGGESTED EXTENSIONS ------------------
+# 1. Log answers to CSV or JSON for further sentiment/topic analysis
+# 2. Add semantic similarity metrics (e.g., cosine distance between embeddings)
+# 3. Build a Streamlit interface or HuggingFace Space for live demo