app.py

import streamlit as st
import textwrap
import pandas as pd
import time
from sentence_transformers import SentenceTransformer, util
from annoy import AnnoyIndex

footer = """
<p style='text-align: center; color: gray;'>Made with inspiration by Abhijeet Singh</p>
"""

shlok_keys = ['Title', 'Chapter', 'Verse', 'Hindi Anuvad' , 'Enlgish Translation']
max_line_length = 100  # Adjust as needed

@st.cache_resource
def load_data():
   hn_filepath = 'Gita.xlsx'
   return pd.read_excel(hn_filepath)


@st.cache_resource
def load_hn_model():
  return SentenceTransformer('all-mpnet-base-v2')

hn_model = load_hn_model()

@st.cache_resource
def build_embeddings(hn_data):
   return [hn_model.encode(hn_data['Enlgish Translation'][i], convert_to_tensor=True).numpy() for i in range(len(hn_data))]

@st.cache_resource
def build_annoy_index(shloka_embeddings):
   embedding_size = len(shloka_embeddings[0])
   annoy_index = AnnoyIndex(embedding_size, metric='angular')
   for i, embedding in enumerate(shloka_embeddings):
       annoy_index.add_item(i, embedding)
   annoy_index.build(18) # 18 trees for faster search
   return annoy_index

def wrap_text(text):
   pass
# st.write("shree ganeshay namah")

hn_data = load_data()



shloka_embeddings = build_embeddings(hn_data)
annoy_index = build_annoy_index(shloka_embeddings)

st.title("GitaShlok Bhagavad Gita Assistant")

st.markdown(footer, unsafe_allow_html=True)

st.markdown(
    """
    <style>
        .reportview-container {
            width: 90%;
        }
    </style>
    """,
    unsafe_allow_html=True
)
st.markdown(
    """
    <style>
        .streamlit-text-container {
            white-space: pre-line;
        }
    </style>
    """,
    unsafe_allow_html=True
)
query = st.text_input("Ask any question related to the Bhagavad Gita: ")

if st.button('Ask'):

    query_embedding = hn_model.encode(query, convert_to_tensor=True).numpy()

    # Use Annoy Index for efficient similarity search
    similar_indices = annoy_index.get_nns_by_vector(query_embedding, 18)

    # Process and display similar Shlokas
    similarities = []
    for curr_index in similar_indices:
        similarity = util.cos_sim(query_embedding, shloka_embeddings[curr_index])
        curr_shlok_details = {key: hn_data[key][curr_index] for key in hn_data}
        similarities.append({"shlok_details": curr_shlok_details, "similarity": similarity})

    # Get the most similar Shloka
    top_result = sorted(similarities, key=lambda x: x["similarity"], reverse=True)[0]
    top_shlok_details = top_result["shlok_details"]
    adhyay_number = top_shlok_details['Chapter'].split(" ")[1]
    shlok_number = top_shlok_details['Verse'].split(" ")[1].split(".")[1]

    st.write("------------------------------")
    st.write(f"{top_shlok_details['Chapter']} , Shloka : {shlok_number}")

    
    
    
    wrapped_text = textwrap.fill(top_shlok_details['Enlgish Translation'], width=max_line_length)
    wrapped_hindi_text=textwrap.fill(top_shlok_details['Hindi Anuvad'], width=max_line_length)

    placeholder = st.empty()

    prev_text=''
    for char in wrapped_text:
      prev_text=prev_text+char
      placeholder.text(prev_text)
      time.sleep(0.01)  # Adjust the sleep duration as needed
    st.write("\n------------------------------")

    hindi_placeholder = st.empty()

    hindi_text=''
    for char in wrapped_hindi_text :
      hindi_text=hindi_text+char
      hindi_placeholder.text(hindi_text)
      time.sleep(0.005)  # Adjust the sleep duration as needed
    st.write("\n------------------------------")