Update app.py

app.py

@@ -1,6 +1,6 @@
 import streamlit as st
 from langchain_google_genai import ChatGoogleGenerativeAI
-from langchain_core.messages import SystemMessage, HumanMessage
+from langchain_core.messages import HumanMessage
 from langchain_core.output_parsers import PydanticOutputParser
 from langchain_core.prompts import PromptTemplate
 from langchain.chains import LLMChain
@@ -17,7 +17,6 @@ from langchain_community.embeddings import HuggingFaceInferenceAPIEmbeddings
 from langchain_community.vectorstores import FAISS
 from langchain_text_splitters import RecursiveCharacterTextSplitter
 
-# Environment Variable Checks
 GOOGLE_API_KEY = os.environ.get("GOOGLE_API_KEY")
 HUGGINGFACE_ACCESS_TOKEN = os.environ.get("HUGGINGFACE_ACCESS_TOKEN")
 
@@ -29,10 +28,10 @@ if not HUGGINGFACE_ACCESS_TOKEN:
     st.error("❌ HUGGINGFACE_ACCESS_TOKEN not found.")
     st.stop()
 
-# Initialize LLM and Embeddings
 llm = ChatGoogleGenerativeAI(
     model="gemini-1.5-pro",
-    google_api_key=GOOGLE_API_KEY
+    google_api_key=GOOGLE_API_KEY,
+    convert_system_message_to_human=True
 )
 
 embeddings = HuggingFaceInferenceAPIEmbeddings(
@@ -40,7 +39,6 @@ embeddings = HuggingFaceInferenceAPIEmbeddings(
     model_name="BAAI/bge-small-en-v1.5"
 )
 
-# Pydantic Models for Structured Output
 class KeyPoint(BaseModel):
     point: str = Field(description="A key point extracted from the document.")
 
@@ -51,10 +49,8 @@ class DocumentAnalysis(BaseModel):
     key_points: List[KeyPoint] = Field(description="List of key points from the document.")
     summary: Summary = Field(description="Summary of the document.")
 
-# Output Parser
 parser = PydanticOutputParser(pydantic_object=DocumentAnalysis)
 
-# Prompt Template
 prompt_template = """
 Analyze the following text and extract key points and a summary.
 {format_instructions}
@@ -66,19 +62,15 @@ prompt = PromptTemplate(
     partial_variables={"format_instructions": parser.get_format_instructions()}
 )
 
-# LLM Chain
 chain = LLMChain(llm=llm, prompt=prompt, output_parser=parser)
 
-# Text Analysis Function
 def analyze_text_structured(text):
     return chain.run(text=text)
 
-# PDF Text Extraction
 def extract_text_from_pdf(pdf_file):
     pdf_reader = PyPDF2.PdfReader(pdf_file)
     return "".join(page.extract_text() for page in pdf_reader.pages)
 
-# JSON to Readable Text
 def json_to_text(analysis):
     text_output = "=== Summary ===\n" + f"{analysis.summary.summary}\n\n"
     text_output += "=== Key Points ===\n"
@@ -86,7 +78,6 @@ def json_to_text(analysis):
         text_output += f"{i}. {key_point.point}\n"
     return text_output
 
-# PDF Report Generation (Updated)
 def create_pdf_report(analysis):
     pdf = FPDF()
     pdf.add_page()
@@ -95,13 +86,11 @@ def create_pdf_report(analysis):
     pdf.cell(200, 10, txt=f"Generated on: {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}", ln=True, align='C')
     pdf.multi_cell(0, 10, txt=json_to_text(analysis))
     
-    # Use BytesIO to create a bytes-like object
     pdf_bytes = io.BytesIO()
     pdf.output(pdf_bytes, dest='S')
     pdf_bytes.seek(0)
     return pdf_bytes.getvalue()
 
-# Word Report Generation (Updated)
 def create_word_report(analysis):
     doc = Document()
     doc.add_heading('PDF Analysis Report', 0)
@@ -109,16 +98,13 @@ def create_word_report(analysis):
     doc.add_heading('Analysis', level=1)
     doc.add_paragraph(json_to_text(analysis))
     
-    # Use BytesIO to create a bytes-like object
     docx_bytes = io.BytesIO()
     doc.save(docx_bytes)
     docx_bytes.seek(0)
     return docx_bytes.getvalue()
 
-# Streamlit Page Configuration
 st.set_page_config(page_title="Chat With PDF", page_icon="📄")
 
-# Custom CSS
 def local_css():
     st.markdown("""
     <style>
@@ -203,7 +189,6 @@ def local_css():
 
 local_css()
 
-# Session State Initialization
 if "current_file" not in st.session_state:
     st.session_state.current_file = None
 if "pdf_summary" not in st.session_state:
@@ -219,14 +204,12 @@ if "vectorstore" not in st.session_state:
 if "messages" not in st.session_state:
     st.session_state.messages = []
 
-# Main App Layout
 st.markdown('<div class="main-header">', unsafe_allow_html=True)
 st.markdown('<div class="flag-stripe"></div>', unsafe_allow_html=True)
 st.title("📄 Chat With PDF")
 st.caption("Your AI-powered Document Analyzer")
 st.markdown('</div>', unsafe_allow_html=True)
 
-# PDF Upload and Analysis Section
 with st.container():
     st.markdown('<div class="card animate-fadeIn">', unsafe_allow_html=True)
     uploaded_file = st.file_uploader("Upload a PDF file", type="pdf")
@@ -276,7 +259,6 @@ with st.container():
                 )
     st.markdown('</div>', unsafe_allow_html=True)
 
-# Document Chat Section
 if "vectorstore" in st.session_state and st.session_state.vectorstore is not None:
     st.subheader("Chat with the Document")
     
@@ -296,8 +278,7 @@ if "vectorstore" in st.session_state and st.session_state.vectorstore is not Non
                 context = "\n".join([doc.page_content for doc in docs])
                 
                 messages = [
-                    SystemMessage(content="You are a helpful assistant. Answer the question based on the provided document context."),
-                    HumanMessage(content=f"Context: {context}\n\nQuestion: {prompt}")
+                    HumanMessage(content=f"You are a helpful assistant. Answer the question based on the provided document context.\n\nContext: {context}\n\nQuestion: {prompt}")
                 ]
                 
                 response = llm.invoke(messages)
@@ -305,7 +286,6 @@ if "vectorstore" in st.session_state and st.session_state.vectorstore is not Non
         
         st.session_state.messages.append({"role": "assistant", "content": response.content})
 
-# Footer
 st.markdown(
     f'<div class="footer">Analysis Time: {st.session_state.analysis_time:.1f}s | Powered by Google Generative AI</div>',
     unsafe_allow_html=True