Update app.py

app.py

@@ -1,5 +1,5 @@
-import os
 import time
+import os
 import streamlit as st
 from langchain_community.vectorstores import FAISS
 from langchain_community.embeddings import HuggingFaceEmbeddings
@@ -7,14 +7,13 @@ from langchain.prompts import PromptTemplate
 from langchain.memory import ConversationBufferWindowMemory
 from langchain.chains import ConversationalRetrievalChain
 from langchain_together import Together
-from footer import footer  # Ensure this module is present in the working directory
-from dotenv import load_dotenv
-import os
 
-# Set Streamlit configuration
-st.set_page_config(page_title="AI Legal App", layout="centered")
+from footer import footer
 
-# Display a logo or banner (replace with a local image or URL)
+# Set the Streamlit page configuration and theme
+st.set_page_config(page_title="BharatLAW", layout="centered")
+
+# Display the logo image
 col1, col2, col3 = st.columns([1, 30, 1])
 with col2:
     st.image("https://github.com/Nike-one/BharatLAW/blob/master/images/banner.png?raw=true", use_column_width=True)
@@ -29,7 +28,7 @@ def hide_hamburger_menu():
 
 hide_hamburger_menu()
 
-# Initialize session state
+# Initialize session state for messages and memory
 if "messages" not in st.session_state:
     st.session_state.messages = []
 
@@ -47,29 +46,44 @@ db_retriever = db.as_retriever(search_type="similarity", search_kwargs={"k": 3})
 
 prompt_template = """
 <s>[INST]
-As a legal chatbot specializing in Indian law, your responses must be concise and accurate:
-- Provide bullet points summarizing key legal aspects.
-- Avoid assumptions or overly specific advice unless requested.
-- Clarify any common misconceptions.
-- Keep responses aligned with general legal principles.
+As a legal chatbot specializing in the Indian Penal Code, you are tasked with providing highly accurate and contextually appropriate responses. Ensure your answers meet these criteria:
+- Respond in a bullet-point format to clearly delineate distinct aspects of the legal query.
+- Each point should accurately reflect the breadth of the legal provision in question, avoiding over-specificity unless directly relevant to the user's query.
+- Clarify the general applicability of the legal rules or sections mentioned, highlighting any common misconceptions or frequently misunderstood aspects.
+- Limit responses to essential information that directly addresses the user's question, providing concise yet comprehensive explanations.
+- Avoid assuming specific contexts or details not provided in the query, focusing on delivering universally applicable legal interpretations unless otherwise specified.
+- Conclude with a brief summary that captures the essence of the legal discussion and corrects any common misinterpretations related to the topic.
+
 CONTEXT: {context}
 CHAT HISTORY: {chat_history}
 QUESTION: {question}
 ANSWER:
+- [Detail the first key aspect of the law, ensuring it reflects general application]
+- [Provide a concise explanation of how the law is typically interpreted or applied]
+- [Correct a common misconception or clarify a frequently misunderstood aspect]
+- [Detail any exceptions to the general rule, if applicable]
+- [Include any additional relevant information that directly relates to the user's query]
 </s>[INST]
 """
 
+
+
 prompt = PromptTemplate(template=prompt_template,
                         input_variables=['context', 'question', 'chat_history'])
-load_dotenv()
+
 api_key = os.getenv('TOGETHER_API_KEY')
 llm = Together(model="mistralai/Mixtral-8x22B-Instruct-v0.1", temperature=0.5, max_tokens=1024, together_api_key=api_key)
 
 qa = ConversationalRetrievalChain.from_llm(llm=llm, memory=st.session_state.memory, retriever=db_retriever, combine_docs_chain_kwargs={'prompt': prompt})
 
 def extract_answer(full_response):
-    """Extracts the assistant's answer from the response."""
-    return full_response.strip()
+    """Extracts the answer from the LLM's full response by removing the instructional text."""
+    answer_start = full_response.find("Response:")
+    if answer_start != -1:
+        answer_start += len("Response:")
+        answer_end = len(full_response)
+        return full_response[answer_start:answer_end].strip()
+    return full_response
 
 def reset_conversation():
     st.session_state.messages = []
@@ -79,28 +93,34 @@ for message in st.session_state.messages:
     with st.chat_message(message["role"]):
         st.write(message["content"])
 
-input_prompt = st.chat_input("Ask your legal query...")
+
+input_prompt = st.chat_input("Say something...")
 if input_prompt:
     with st.chat_message("user"):
         st.markdown(f"**You:** {input_prompt}")
 
     st.session_state.messages.append({"role": "user", "content": input_prompt})
     with st.chat_message("assistant"):
-        with st.spinner("Analyzing..."):
+        with st.spinner("Thinking 💡..."):
             result = qa.invoke(input=input_prompt)
             message_placeholder = st.empty()
             answer = extract_answer(result["answer"])
 
-            # Simulated typing effect
-            response = ""
-            for char in answer:
-                response += char
-                time.sleep(0.02)
-                message_placeholder.markdown(response + " |", unsafe_allow_html=True)
+            # Initialize the response message
+            full_response = "⚠️ **_Gentle reminder: We generally ensure precise information, but do double-check._** \n\n\n"
+            for chunk in answer:
+                # Simulate typing by appending chunks of the response over time
+                full_response += chunk
+                time.sleep(0.02)  # Adjust the sleep time to control the "typing" speed
+                message_placeholder.markdown(full_response + " |", unsafe_allow_html=True)
 
         st.session_state.messages.append({"role": "assistant", "content": answer})
 
-        if st.button('🗑️ Reset Chat', on_click=reset_conversation):
+        if st.button('🗑️ Reset All Chat', on_click=reset_conversation):
             st.experimental_rerun()
 
+
+
+# Define the CSS to style the footer
 footer()
+