Update app.py

app.py

@@ -372,12 +372,42 @@ def respond(message, history, model, temperature, num_calls, use_web_search, sel
 
 logging.basicConfig(level=logging.DEBUG)
 
-INSTRUCTION_PROMPTS = {
-    "Asset Managers": "Summarize the key financial metrics, assets under management, and performance highlights for this asset management company.",
-    "Consumer Finance Companies": "Provide a summary of the company's loan portfolio, interest income, credit quality, and key operational metrics.",
-    "Mortgage REITs": "Summarize the REIT's mortgage-backed securities portfolio, net interest income, book value per share, and dividend yield.",
-    # Add more instruction prompts as needed
-}
+def get_context_for_summary(selected_docs):
+    embed = get_embeddings()
+    if os.path.exists("faiss_database"):
+        database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
+        retriever = database.as_retriever(search_kwargs={"k": 5})  # Retrieve top 5 most relevant chunks
+        
+        # Create a generic query that covers common financial summary topics
+        generic_query = "financial performance revenue profit assets liabilities cash flow key metrics highlights"
+        
+        relevant_docs = retriever.get_relevant_documents(generic_query)
+        filtered_docs = [doc for doc in relevant_docs if doc.metadata["source"] in selected_docs]
+        
+        if not filtered_docs:
+            return "No relevant information found in the selected documents for summary generation."
+        
+        context_str = "\n".join([doc.page_content for doc in filtered_docs])
+        return context_str
+    else:
+        return "No documents available for summary generation."
+
+def get_context_for_query(query, selected_docs):
+    embed = get_embeddings()
+    if os.path.exists("faiss_database"):
+        database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
+        retriever = database.as_retriever(search_kwargs={"k": 3})  # Retrieve top 3 most relevant chunks
+        
+        relevant_docs = retriever.get_relevant_documents(query)
+        filtered_docs = [doc for doc in relevant_docs if doc.metadata["source"] in selected_docs]
+        
+        if not filtered_docs:
+            return "No relevant information found in the selected documents for the given query."
+        
+        context_str = "\n".join([doc.page_content for doc in filtered_docs])
+        return context_str
+    else:
+        return "No documents available to answer the query."
 
 def get_response_from_cloudflare(prompt, context, query, num_calls=3, temperature=0.2, search_type="pdf"):
     headers = {
@@ -465,9 +495,12 @@ After writing the document, please provide a list of sources used in your respon
                     main_content += chunk
                     yield main_content, ""  # Yield partial main content without sources
 
-from langchain_community.vectorstores import FAISS
-from langchain_core.vectorstores import VectorStore
-from langchain_core.documents import Document
+INSTRUCTION_PROMPTS = {
+    "Asset Managers": "Summarize the key financial metrics, assets under management, and performance highlights for this asset management company.",
+    "Consumer Finance Companies": "Provide a summary of the company's loan portfolio, interest income, credit quality, and key operational metrics.",
+    "Mortgage REITs": "Summarize the REIT's mortgage-backed securities portfolio, net interest income, book value per share, and dividend yield.",
+    # Add more instruction prompts as needed
+}
 
 def get_response_from_pdf(query, model, selected_docs, num_calls=3, temperature=0.2):
     logging.info(f"Entering get_response_from_pdf with query: {query}, model: {model}, selected_docs: {selected_docs}")
@@ -557,43 +590,6 @@ css = """
 }
 """
 
-def get_context_for_summary(selected_docs):
-    embed = get_embeddings()
-    if os.path.exists("faiss_database"):
-        database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
-        retriever = database.as_retriever(search_kwargs={"k": 5})  # Retrieve top 5 most relevant chunks
-        
-        # Create a generic query that covers common financial summary topics
-        generic_query = "financial performance revenue profit assets liabilities cash flow key metrics highlights"
-        
-        relevant_docs = retriever.get_relevant_documents(generic_query)
-        filtered_docs = [doc for doc in relevant_docs if doc.metadata["source"] in selected_docs]
-        
-        if not filtered_docs:
-            return "No relevant information found in the selected documents for summary generation."
-        
-        context_str = "\n".join([doc.page_content for doc in filtered_docs])
-        return context_str
-    else:
-        return "No documents available for summary generation."
-
-def get_context_for_query(query, selected_docs):
-    embed = get_embeddings()
-    if os.path.exists("faiss_database"):
-        database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
-        retriever = database.as_retriever(search_kwargs={"k": 3})  # Retrieve top 3 most relevant chunks
-        
-        relevant_docs = retriever.get_relevant_documents(query)
-        filtered_docs = [doc for doc in relevant_docs if doc.metadata["source"] in selected_docs]
-        
-        if not filtered_docs:
-            return "No relevant information found in the selected documents for the given query."
-        
-        context_str = "\n".join([doc.page_content for doc in filtered_docs])
-        return context_str
-    else:
-        return "No documents available to answer the query."
-
 uploaded_documents = []
 
 def display_documents():
@@ -603,24 +599,23 @@ def display_documents():
         label="Select documents to query or delete"
     )
 
-# Add this new function
-def refresh_documents():
-    global uploaded_documents
-    uploaded_documents = load_documents()
-    return display_documents()
-
 def initial_conversation():
     return [
         (None, "Welcome! I'm your AI assistant for web search and PDF analysis. Here's how you can use me:\n\n"
                 "1. Set the toggle for Web Search and PDF Search from the checkbox in Additional Inputs drop down window\n"
                 "2. Use web search to find information\n"
-                "3. Ask questions about uploaded PDF documents\n"
-                "4. Generate summaries for specific entity types\n\n"
+                "3. Upload the documents and ask questions about uploaded PDF documents by selecting your respective document\n"
+                "4. For any queries feel free to reach out @[email protected] or discord - shreyas094\n\n"
                 "To get started, upload some PDFs or ask me a question!")
     ]
+# Add this new function
+def refresh_documents():
+    global uploaded_documents
+    uploaded_documents = load_documents()
+    return display_documents()
 
 # Define the checkbox outside the demo block
-document_selector = display_documents()
+document_selector = gr.CheckboxGroup(label="Select documents to query")
 
 use_web_search = gr.Checkbox(label="Use Web Search", value=True)
 
@@ -678,7 +673,7 @@ demo = gr.ChatInterface(
 
 # Add file upload functionality
 with demo:
-    gr.Markdown("## Upload PDF Documents")
+    gr.Markdown("## Upload and Manage PDF Documents")
 
     with gr.Row():
         file_input = gr.Files(label="Upload your PDF documents", file_types=[".pdf"])