import streamlit as st from io import BytesIO from datetime import datetime Assuming…
- User
import streamlit as st from io import BytesIO from datetime import datetime
Assuming these are correctly defined and accessible
from utils.pdf_utils import extract_text from clients.azure_client import AzureClient from clients.gemini_client import GeminiClient
from agents.tariff_agent import TariffTradeAgent from agents.sales_agent import SalesAgent from agents.supply_chain_agent import SupplyChainAgent from agents.overall_agent import OverallAgent from agents.enhanced_chat_agent import EnhancedChatAgent from agents.qa_agent import QAAgent # Add QA agent import
def analyze_documents(mode, agent, processed_files, sess, custom_prompt=None):
# Process individual files for filename, file_bytes in sess['uploaded_files'].items(): if filename not in processed_files: try: file_obj = BytesIO(file_bytes) file_obj.name = filename text = extract_text(file_obj) if text: if mode == "overall": topics = agent.analyze_document(text, custom_prompt) elif mode == "qa": # QA mode uses analyze_question instead qa_content = agent.analyze_question(text, custom_prompt) processed_files[filename] = { "text": text, "qa": qa_content, # Store QA content directly } else: topics = agent.analyze_document(text, custom_prompt) processed_files[filename] = { "text": text, "topics": topics, "context": text, } st.toast(f"✅ Processed {filename}") else: st.error(f"❌ Failed to extract text for {filename}") except Exception as e: st.error(f"Error processing {filename}: {str(e)}") # Generate consolidated summary (skip for QA) if mode != "qa": summary = agent.generate_consolidated_summary_from_topics(processed_files, custom_prompt) if mode == "overall": sess['consolidated_summary_overall'] = summary elif mode == "tariff": sess['consolidated_summary_tariff'] = summary elif mode == "sales": sess['consolidated_summary_sales'] = summary elif mode == "supplychain": sess['consolidated_summary_supplychain'] = summary sess['last_analysis_mode'] = mode sess['active_file'] = None sess['active_topic'] = None st.success(f"Analysis and summary generated for {mode.replace('supplychain', 'supply chain')}.")def main(): st.set_page_config(page_title="Macy's Document Analysis Pro", layout="wide") st.title("📄 Macy's Document Analysis & Chat Tool")
# Mode selection provider_select = st.radio("Select AI Provider", ["Azure GPT", "Google Gemini"]) # Session Initialization if "session" not in st.session_state: st.session_state.session = { 'uploaded_files': {}, 'processed_files_overall': {}, 'processed_files_tariff': {}, 'processed_files_sales': {}, 'processed_files_supplychain': {}, 'processed_files_qa': {}, # New QA processed files 'active_file': None, 'active_topic': None, 'topic_details_cache': {}, 'chat_history': [], 'consolidated_summary_overall': None, 'consolidated_summary_tariff': None, 'consolidated_summary_sales': None, 'consolidated_summary_supplychain': None, 'last_analysis_mode': None, 'custom_prompt_enabled': {}, 'custom_prompt_text': {} } if "client" not in st.session_state or st.session_state.get("provider") != provider_select: if provider_select == "Azure GPT": st.session_state.client = AzureClient() is_azure = True else: st.session_state.client = GeminiClient() is_azure = False # Initialize agents (add QA agent) st.session_state.agent_overall = OverallAgent(st.session_state.client, is_azure) st.session_state.agent_tariff = TariffTradeAgent(st.session_state.client, is_azure) st.session_state.agent_sales = SalesAgent(st.session_state.client, is_azure) st.session_state.agent_supplychain = SupplyChainAgent(st.session_state.client, is_azure) st.session_state.agent_qa = QAAgent(st.session_state.client, is_azure) # New QA agent st.session_state.chat_agent = EnhancedChatAgent(st.session_state.client, is_azure) st.session_state.provider = provider_select sess = st.session_state.session # --- File Upload --- st.subheader("1. Upload Documents") uploaded_files = st.file_uploader( "Upload PDF documents", type=["pdf"], accept_multiple_files=True, key="file_uploader", ) if uploaded_files: new_files_added = False for file in uploaded_files: if file.name not in sess['uploaded_files']: sess['uploaded_files'][file.name] = file.getvalue() new_files_added = True if new_files_added: with st.spinner("🔍 Analyzing new documents with Overall Agent..."): # Always run initial overall analysis for new files analyze_documents( mode="overall", agent=st.session_state.agent_overall, processed_files=sess['processed_files_overall'], sess=sess ) # --- Custom Prompts & Analysis --- st.subheader("2. Custom Prompts & Analysis") # Initialize custom prompts in session state if 'custom_prompts' not in sess: sess['custom_prompts'] = { 'overall': { 'summary': OverallAgent.prompt_summary, 'analyze': OverallAgent.prompt_analyze, 'detail': OverallAgent.prompt_detail }, 'tariff': { 'summary': TariffTradeAgent.prompt_summary, 'analyze': TariffTradeAgent.prompt_analyze, 'detail': TariffTradeAgent.prompt_detail }, 'sales': { 'summary': SalesAgent.prompt_summary, 'analyze': SalesAgent.prompt_analyze, 'detail': SalesAgent.prompt_detail }, 'supplychain': { 'summary': SupplyChainAgent.prompt_summary, 'analyze': SupplyChainAgent.prompt_analyze, 'detail': SupplyChainAgent.prompt_detail }, 'qa': { # New QA prompts 'analyze': QAAgent.prompt_analyze } } # Create tabs for each agent's custom prompts (add QA tab) tab_overall, tab_tariff, tab_sales, tab_supply, tab_qa = st.tabs([ "📑 Overall", "🔍 Tariff", "📈 Sales", "🚚 Supply Chain", "❓ Q&A" ]) # Store which mode we're generating for generate_mode = None with tab_overall: st.subheader("Overall Prompts") use_custom = st.checkbox("✏️ Use Custom Prompts", key="custom_overall") if use_custom: sess['custom_prompts']['overall']['summary'] = st.text_area( "Summary Prompt:", value=sess['custom_prompts']['overall']['summary'], height=150, key="overall_prompt_summary" ) sess['custom_prompts']['overall']['analyze'] = st.text_area( "Analysis Prompt:", value=sess['custom_prompts']['overall']['analyze'], height=150, key="overall_prompt_analyze" ) sess['custom_prompts']['overall']['detail'] = st.text_area( "Detail Prompt:", value=sess['custom_prompts']['overall']['detail'], height=150, key="overall_prompt_detail" ) if st.button("🔄 Generate with Custom Prompts", key="generate_overall"): generate_mode = "overall" with tab_tariff: st.subheader("Tariff Prompts") use_custom = st.checkbox("✏️ Use Custom Prompts", key="custom_tariff") if use_custom: sess['custom_prompts']['tariff']['summary'] = st.text_area( "Summary Prompt:", value=sess['custom_prompts']['tariff']['summary'], height=150, key="tariff_prompt_summary" ) sess['custom_prompts']['tariff']['analyze'] = st.text_area( "Analysis Prompt:", value=sess['custom_prompts']['tariff']['analyze'], height=150, key="tariff_prompt_analyze" ) sess['custom_prompts']['tariff']['detail'] = st.text_area( "Detail Prompt:", value=sess['custom_prompts']['tariff']['detail'], height=150, key="tariff_prompt_detail" ) if st.button("🔄 Generate with Custom Prompts", key="generate_tariff"): generate_mode = "tariff" with tab_sales: st.subheader("Sales Prompts") use_custom = st.checkbox("✏️ Use Custom Prompts", key="custom_sales") if use_custom: sess['custom_prompts']['sales']['summary'] = st.text_area( "Summary Prompt:", value=sess['custom_prompts']['sales']['summary'], height=150, key="sales_prompt_summary" ) sess['custom_prompts']['sales']['analyze'] = st.text_area( "Analysis Prompt:", value=sess['custom_prompts']['sales']['analyze'], height=150, key="sales_prompt_analyze" ) sess['custom_prompts']['sales']['detail'] = st.text_area( "Detail Prompt:", value=sess['custom_prompts']['sales']['detail'], height=150, key="sales_prompt_detail" ) if st.button("🔄 Generate with Custom Prompts", key="generate_sales"): generate_mode = "sales" with tab_supply: st.subheader("Supply Chain Prompts") use_custom = st.checkbox("✏️ Use Custom Prompts", key="custom_supply") if use_custom: sess['custom_prompts']['supplychain']['summary'] = st.text_area( "Summary Prompt:", value=sess['custom_prompts']['supplychain']['summary'], height=150, key="supply_prompt_summary" ) sess['custom_prompts']['supplychain']['analyze'] = st.text_area( "Analysis Prompt:", value=sess['custom_prompts']['supplychain']['analyze'], height=150, key="supply_prompt_analyze" ) sess['custom_prompts']['supplychain']['detail'] = st.text_area( "Detail Prompt:", value=sess['custom_prompts']['supplychain']['detail'], height=150, key="supply_prompt_detail" ) if st.button("🔄 Generate with Custom Prompts", key="generate_supplychain"): generate_mode = "supplychain" # New QA tab with tab_qa: st.subheader("Q&A Extraction Prompts") use_custom = st.checkbox("✏️ Use Custom Prompts", key="custom_qa") if use_custom: sess['custom_prompts']['qa']['analyze'] = st.text_area( "Q&A Extraction Prompt:", value=sess['custom_prompts']['qa']['analyze'], height=300, key="qa_prompt_analyze" ) if st.button("🔄 Generate with Custom Prompts", key="generate_qa"): generate_mode = "qa" # --- Analysis Buttons --- st.subheader("Quick Analysis") col0, col1, col2, col3, col4 = st.columns(5) # Add extra column for QA analyze_overall = col0.button("📑 Overall Summary") analyze_tariff = col1.button("🔍 Analyze Tariffs") analyze_sales = col2.button("📈 Analyze Sales") analyze_supplychain = col3.button("🚚 Supply Chain") analyze_qa = col4.button("❓ Extract Q&A") # New QA button # Handle generation requests if generate_mode or analyze_overall or analyze_tariff or analyze_sales or analyze_supplychain or analyze_qa: if not sess['uploaded_files']: st.error("Please upload documents first.") st.stop() mode = generate_mode or ( "overall" if analyze_overall else "tariff" if analyze_tariff else "sales" if analyze_sales else "supplychain" if analyze_supplychain else "qa" if analyze_qa else None ) agent = None processed_files = None if mode == "overall": agent = st.session_state.agent_overall processed_files = sess['processed_files_overall'] elif mode == "tariff": agent = st.session_state.agent_tariff processed_files = sess['processed_files_tariff'] elif mode == "sales": agent = st.session_state.agent_sales processed_files = sess['processed_files_sales'] elif mode == "supplychain": agent = st.session_state.agent_supplychain processed_files = sess['processed_files_supplychain'] elif mode == "qa": # New QA mode agent = st.session_state.agent_qa processed_files = sess['processed_files_qa'] # Update agent with custom prompts if used if generate_mode: if mode != "qa": agent.prompt_summary = sess['custom_prompts'][mode]['summary'] agent.prompt_analyze = sess['custom_prompts'][mode]['analyze'] agent.prompt_detail = sess['custom_prompts'][mode]['detail'] else: agent.prompt_analyze = sess['custom_prompts'][mode]['analyze'] st.toast(f"Using custom prompts for {mode} analysis") else: # Reset to default prompts if mode == "overall": agent.prompt_summary = OverallAgent.prompt_summary agent.prompt_analyze = OverallAgent.prompt_analyze agent.prompt_detail = OverallAgent.prompt_detail elif mode == "tariff": agent.prompt_summary = TariffTradeAgent.prompt_summary agent.prompt_analyze = TariffTradeAgent.prompt_analyze agent.prompt_detail = TariffTradeAgent.prompt_detail elif mode == "sales": agent.prompt_summary = SalesAgent.prompt_summary agent.prompt_analyze = SalesAgent.prompt_analyze agent.prompt_detail = SalesAgent.prompt_detail elif mode == "supplychain": agent.prompt_summary = SupplyChainAgent.prompt_summary agent.prompt_analyze = SupplyChainAgent.prompt_analyze agent.prompt_detail = SupplyChainAgent.prompt_detail elif mode == "qa": agent.prompt_analyze = QAAgent.prompt_analyze with st.spinner(f"🔍 Analyzing {mode.replace('supplychain', 'supply chain').replace('qa', 'Q&A')} documents..."): analyze_documents( mode, agent, processed_files, sess ) # --- Main Display --- last_mode = sess['last_analysis_mode'] if last_mode is None: st.info("Select an analysis type to begin.") return # Consolidated data setup (add QA mode) mode_config = { 'overall': { 'files': sess['processed_files_overall'], 'summary': sess['consolidated_summary_overall'], 'sidebar_header': "Overall Topics", 'main_header': "🌟 Executive Summary", 'agent': st.session_state.agent_overall }, 'tariff': { 'files': sess['processed_files_tariff'], 'summary': sess['consolidated_summary_tariff'], 'sidebar_header': "Tariff Topics", 'main_header': "🌟 Tariff Analysis", 'agent': st.session_state.agent_tariff }, 'sales': { 'files': sess['processed_files_sales'], 'summary': sess['consolidated_summary_sales'], 'sidebar_header': "Sales Topics", 'main_header': "🌟 Sales Analysis", 'agent': st.session_state.agent_sales }, 'supplychain': { 'files': sess['processed_files_supplychain'], 'summary': sess['consolidated_summary_supplychain'], 'sidebar_header': "Supply Chain Topics", 'main_header': "🌟 Supply Chain Analysis", 'agent': st.session_state.agent_supplychain }, 'qa': { # New QA config 'files': sess['processed_files_qa'], 'summary': None, # QA doesn't have consolidated summary 'sidebar_header': "Documents with Q&A", 'main_header': "❓ Extracted Q&A", 'agent': st.session_state.agent_qa } } current_mode = mode_config[last_mode] # --- Sidebar Topics --- st.sidebar.header(current_mode['sidebar_header']) if last_mode == "overall": topics_by_file = {} for filename, filedata in current_mode['files'].items(): for topic in filedata["topics"]: topics_by_file.setdefault(topic['name'], []).append({ 'file': filename, 'score': topic['score'] }) a = sorted(topics_by_file.items(), key=lambda x: max(x[1], key=lambda y: y['score'])['score'], reverse=True) before_view_more , after_view_more = a[:3] , a[3:] for topic_name, occurrences in before_view_more: total_score = sum(occ['score'] for occ in occurrences) file_count = len(occurrences) btn_label = f"{topic_name} (📈 {total_score:.1f} | 📑 {file_count})" if st.sidebar.button( btn_label, key=f"overall_topic_{topic_name}", use_container_width=True, ): sess['active_topic'] = topic_name sess['active_file'] = None if st.sidebar.button('View More',key = '1234' , use_container_width=True): for topic_name, occurrences in after_view_more: total_score = sum(occ['score'] for occ in occurrences) file_count = len(occurrences) btn_label = f"{topic_name} (📈 {total_score:.1f} | 📑 {file_count})" if st.sidebar.button( btn_label, key=f"overall_topic_more_{topic_name}", # Changed key to avoid collision use_container_width=True, ): sess['active_topic'] = topic_name sess['active_file'] = None elif last_mode == "qa": # New QA sidebar for filename in current_mode['files']: if st.sidebar.button( f"📄 {filename}", key=f"qa_file_{filename}", use_container_width=True, ): sess['active_file'] = filename sess['active_topic'] = None else: # For Tariff, Sales, Supply Chain modes for filename, filedata in current_mode['files'].items(): expanded = (filename == sess['active_file']) with st.sidebar.expander(f"📄 {filename}", expanded=expanded): topics = sorted(filedata["topics"], key=lambda t: t["score"], reverse=True) for topic in topics[:3]: is_selected = ( sess['active_file'] == filename and sess['active_topic'] == topic['name'] ) style = "primary" if is_selected else "secondary" if st.button( f"• {topic['name']} (Score: {topic['score']})", key=f"{last_mode}_topic_{filename}_{topic['name']}", type=style, use_container_width=True, ): sess['active_file'] = filename sess['active_topic'] = topic['name'] if len(topics) > 3 and st.sidebar.button(f'View More for {filename}',key = f'{filename}_view_more' ,use_container_width=True): for topic in topics[3:]: is_selected = ( sess['active_file'] == filename and sess['active_topic'] == topic['name'] ) style = "primary" if is_selected else "secondary" if st.button( f"• {topic['name']} (Score: {topic['score']})", key=f"{last_mode}_topic_more_{filename}_{topic['name']}", # Changed key type=style, use_container_width=True, ): sess['active_file'] = filename sess['active_topic'] = topic['name'] # --- Topic Analysis --- st.subheader("Detailed Analysis") if last_mode == "overall" and sess['active_topic']: mentions = [] for filename, filedata in current_mode['files'].items(): for topic in filedata["topics"]: if topic['name'] == sess['active_topic']: mentions.append({ 'file': filename, 'summary': topic.get('summary', ''), 'score': topic['score'], 'context': filedata['context'] }) with st.expander("📊 Topic Prevalence"): cols = st.columns(3) cols[0].metric("Total Documents", len(mentions)) avg_relevance = (sum(m['score'] for m in mentions)/len(mentions)) if mentions else 0 cols[1].metric("Average Relevance", f"{avg_relevance:.1f}") cols[2].metric("First Mentioned", mentions[0]['file'] if mentions else "N/A") if mentions: combined_context = "\n\n".join(f"=== {m['file']} ===\n{m['context']}" for m in mentions) key_cache = (last_mode, sess['active_topic']) if key_cache in sess['topic_details_cache']: st.markdown(sess['topic_details_cache'][key_cache]) else: placeholder = st.empty() collected_text = "" try: for chunk in current_mode['agent'].generate_topic_detail(combined_context, sess['active_topic']): collected_text = chunk placeholder.markdown(collected_text + (" ▌" if "▌" in collected_text else "")) sess['topic_details_cache'][key_cache] = collected_text except Exception as e: st.error(f"Analysis error: {str(e)}") elif last_mode == "qa" and sess['active_file']: # New QA display filename = sess['active_file'] qa_content = current_mode['files'][filename].get('qa', '') if qa_content: st.markdown(qa_content) else: st.info("No Q&A content found for this document") elif sess['active_file'] and sess['active_topic']: key_cache = (last_mode, sess['active_file'], sess['active_topic']) if key_cache in sess['topic_details_cache']: st.markdown(sess['topic_details_cache'][key_cache]) else: placeholder = st.empty() collected_text = "" try: for chunk in current_mode['agent'].generate_topic_detail( current_mode['files'][sess['active_file']]['context'], sess['active_topic'], ): collected_text = chunk placeholder.markdown(collected_text + (" ▌" if "▌" in collected_text else "")) sess['topic_details_cache'][key_cache] = collected_text except Exception as e: st.error(f"Analysis error: {str(e)}") else: st.info("Select a topic from the sidebar") st.markdown("---") # --- Consolidated Summary --- # Skip for QA mode if last_mode != "qa": st.header(current_mode['main_header']) if current_mode['summary']: st.markdown(current_mode['summary']) else: st.info("Summary not available yet") st.divider() # --- Chat Interface --- st.subheader("💬 Document Chat") # File selection available_files = list(current_mode['files'].keys()) selected_files = st.multiselect( "Select documents for context:", available_files, key=f"chat_selector_{last_mode}" ) # Chat history for msg in sess['chat_history']: with st.chat_message(msg["role"]): st.markdown(msg["content"]) # Question handling question = st.chat_input("Ask about the documents...") if question: if not selected_files: st.error("Please select documents first") st.stop() sess['chat_history'].append({ "role": "user", "content": question, "timestamp": datetime.utcnow().isoformat(), }) try: context = "\n\n".join([ f"=== {fname} ===\n{current_mode['files'][fname]['context']}" for fname in selected_files ]) with st.chat_message("assistant"): response_placeholder = st.empty() full_response = "" q_type = None # Ensure custom_prompt_text_for_mode is available here if needed for chat_agent # For now, it's used only for analysis in analyze_documents, adjust if chat_agent needs it. for chunk, q_type in st.session_state.chat_agent.generate_answer(question, context): full_response = chunk response_placeholder.markdown(full_response + " ▌") response_placeholder.markdown(full_response) sess['chat_history'].append({ "role": "assistant", "content": full_response, "timestamp": datetime.utcnow().isoformat(), "analysis_type": q_type }) except Exception as e: st.error(f"Chat error: {str(e)}") sess['chat_history'].append({ "role": "assistant", "content": f"Error: {str(e)}", "timestamp": datetime.utcnow().isoformat(), }) # Reset button if st.button("🔄 Reset Session", type="secondary"): # Preserve long-lived objects like client and agents to avoid overhead on re-init preserved_keys = [ "client", "agent_overall", "agent_tariff", "agent_sales", "agent_supplychain", "agent_qa", "chat_agent", "provider" ] preserved = {k: st.session_state[k] for k in preserved_keys if k in st.session_state} # Clear entire session state st.session_state.clear() # Restore preserved keys back st.session_state.update(preserved) # Reset session-specific stored data completely st.session_state.session = { 'uploaded_files': {}, 'processed_files_overall': {}, 'processed_files_tariff': {}, 'processed_files_sales': {}, 'processed_files_supplychain': {}, 'processed_files_qa': {}, # New QA processed files 'active_file': None, 'active_topic': None, 'topic_details_cache': {}, 'chat_history': [], 'consolidated_summary_overall': None, 'consolidated_summary_tariff': None, 'consolidated_summary_sales': None, 'consolidated_summary_supplychain': None, 'last_analysis_mode': None, 'custom_prompt_enabled': {}, 'custom_prompt_text': {} } # Trigger a rerun for fresh UI without old content st.rerun()if name == "main": main()
- Assistant
Your code shows a fully integrated Streamlit app for document analysis using multiple specialized agents, including a newly introduced QA agent mode. It adds a new Q&A extraction functionality alongside your existing Overall, Tariff, Sales, and Supply Chain analysis capabilities.
Here are some comments and suggestions to ensure everything works smoothly and to improve maintainability:
1. QA Integration Looks Good
- You added
QAAgentwith its corresponding processed files dictprocessed_files_qain session. - Created a separate tab with its own custom prompt.
- Added a dedicated sidebar to list documents for Q&A and show extracted Q&A.
- Updated analysis triggering and agent selection logic.
- Skipped summary display for QA as expected.
2. Minor Considerations & Possible Improvements
a) Storing QA results in
processed_files[filename]Currently, in
analyze_documents()under QA mode:processed_files[filename] = { "text": text, "qa": qa_content, # Store QA content directly }- Make sure
qa_contentis a string or markdown-friendly content. - Your display code shows it via
st.markdown(qa_content), so the format is probably okay. - If
qa_contentis rather complex (like a dict or list), consider serializing it properly or formatting into markdown.
b) For custom prompt toggles in UI tabs:
- You track whether to use custom prompts only via checkboxes like
"custom_overall". But for mode QA,sess['custom_prompts']['qa']relies only on a singleanalyzeprompt. - That is consistent with how other agents operate, just keep it documented.
c) Chat with multiple modes and custom prompts
- You reset agent prompts depending on whether analysis is initiated via button or generate with custom prompt.
- Chat agent uses
st.session_state.chat_agent.generate_answer()independently. - If custom prompts ever need to influence chat, consider handling that via
st.session_state.custom_prompt_textperhaps. - Currently, chat interface does not allow switching between agents. It always uses
st.session_state.chat_agent. - You might want to allow chat about different modes, or be explicit in the UI about what chat is for.
d) Sidebar's "View More" buttons for topics and topics lists
- You used keys like
'1234'for "View More" in the overall topics sidebar. - For better clarity and to avoid key collision, consider descriptive keys like
'overall_view_more'.
e) Exception handling
- You catch exceptions well during file processing and topic detail generation.
- Consider optionally logging exceptions for easier debugging.
f) Imported utils and clients
- Make sure your imports for
extract_text,AzureClient,GeminiClient, agents are all working and your environment has all dependencies.
g) Reset Session Button
- Preserving client and agents is smart to avoid reload overhead.
- Just ensure that if these objects hold large state that is best refreshed on reset, you might want to reinstantiate them.
3. Potential Future Enhancements
- Pagination or search in topic lists if topics are large.
- Enable chat to switch agents or specify domain (Overall, Tariff, QA).
- Save user chat history across sessions, export logs.
- Add document download/export functionality for extracted Q&A or analysis results.
- Add progress bars or early feedback for long-running analysis.
Summary
Your code is well structured and integrates QA functionality elegantly with the existing framework. You support uploading multiple PDFs, customizable prompts, multi-agent analysis, and Q&A extraction.
If you want, I can help with code snippets or refactoring suggestions. Or help with tests, debugging, or UI/UX enhancements. Just ask!
- You added