Upload 6 files

app.py

@@ -1,1013 +1,1329 @@
-"""
-ConversAI - AI-Powered Qualitative Research Assistant
-Production-grade survey generation, translation, and analysis platform
-"""
-import gradio as gr
-import json
-import os
-import traceback
-from typing import Dict, List, Optional
-
-from llm_backend import LLMBackend, LLMProvider
-from survey_generator import SurveyGenerator
-from survey_translator import SurveyTranslator
-from data_analyzer import DataAnalyzer
-from export_utils import (save_json_file, survey_to_csv, analysis_to_markdown_file,
-                          conversation_to_transcript, conversation_to_json, conversation_to_csv,
-                          flow_to_markdown)
-from conversation_flow import ConversationFlow, ConversationNode, create_example_flow
-from conversation_session import ConversationSession, SessionManager
-from conversation_moderator import ConversationModerator
-
-
-# Global state for current survey
-current_survey = None
-current_responses = []
-
-# Global state for conversational research
-current_flow = None
-session_manager = SessionManager()
-current_session = None
-saved_flows = {}
-
-
-def initialize_backend():
-    """Initialize LLM backend based on environment"""
-    try:
-        # Debug: Print all environment variables related to LLM
-        print("=== LLM Backend Initialization ===")
-        print(f"HF_TOKEN: {'SET' if os.getenv('HF_TOKEN') else 'NOT SET'}")
-        print(f"HUGGINGFACE_API_KEY: {'SET' if os.getenv('HUGGINGFACE_API_KEY') else 'NOT SET'}")
-        print(f"OPENAI_API_KEY: {'SET' if os.getenv('OPENAI_API_KEY') else 'NOT SET'}")
-        print(f"ANTHROPIC_API_KEY: {'SET' if os.getenv('ANTHROPIC_API_KEY') else 'NOT SET'}")
-        print(f"LLM_PROVIDER: {os.getenv('LLM_PROVIDER', 'NOT SET')}")
-
-        # Check for explicit provider setting
-        provider_env = os.getenv("LLM_PROVIDER", "").lower()
-
-        # Priority 1: Explicitly set provider
-        if provider_env == "openai" and os.getenv("OPENAI_API_KEY"):
-            print("Using OpenAI (explicit)")
-            return LLMBackend(provider=LLMProvider.OPENAI)
-        elif provider_env == "anthropic" and os.getenv("ANTHROPIC_API_KEY"):
-            print("Using Anthropic (explicit)")
-            return LLMBackend(provider=LLMProvider.ANTHROPIC)
-        elif provider_env == "huggingface" and (os.getenv("HUGGINGFACE_API_KEY") or os.getenv("HF_TOKEN")):
-            api_key = os.getenv("HUGGINGFACE_API_KEY") or os.getenv("HF_TOKEN")
-            print("Using HuggingFace (explicit)")
-            return LLMBackend(provider=LLMProvider.HUGGINGFACE, api_key=api_key)
-        elif provider_env == "lm_studio":
-            print("Using LM Studio (explicit)")
-            return LLMBackend(provider=LLMProvider.LM_STUDIO)
-
-        # Priority 2: Auto-detect based on available credentials
-        # HF_TOKEN is automatically available in HF Spaces, so check it first
-        hf_token = os.getenv("HF_TOKEN") or os.getenv("HUGGINGFACE_API_KEY")
-        if hf_token:
-            print(f"Auto-detected HuggingFace credentials, using HF Inference API")
-            print(f"Token preview: {hf_token[:10]}...")
-            return LLMBackend(provider=LLMProvider.HUGGINGFACE, api_key=hf_token)
-        elif os.getenv("OPENAI_API_KEY"):
-            print(f"Auto-detected OpenAI credentials")
-            return LLMBackend(provider=LLMProvider.OPENAI)
-        elif os.getenv("ANTHROPIC_API_KEY"):
-            print(f"Auto-detected Anthropic credentials")
-            return LLMBackend(provider=LLMProvider.ANTHROPIC)
-        else:
-            # No credentials found - return None to show error in UI
-            print("="*60)
-            print("WARNING: No LLM provider credentials found!")
-            print("="*60)
-            print("For HuggingFace Spaces:")
-            print("  - HF_TOKEN should be automatically available")
-            print("  - Make sure your Space is PUBLIC")
-            print("  - Or add HUGGINGFACE_API_KEY in Settings")
-            print("")
-            print("For other providers, set one of:")
-            print("  - OPENAI_API_KEY")
-            print("  - ANTHROPIC_API_KEY")
-            print("  - HUGGINGFACE_API_KEY")
-            print("="*60)
-            return None
-
-    except Exception as e:
-        print(f"Error during backend initialization: {e}")
-        import traceback
-        traceback.print_exc()
-        return None
-
-
-# Initialize components
-llm_backend = initialize_backend()
-
-# Only initialize if backend is available
-if llm_backend:
-    survey_gen = SurveyGenerator(llm_backend)
-    survey_trans = SurveyTranslator(llm_backend)
-    data_analyzer = DataAnalyzer(llm_backend)
-    print(f"✓ ConversAI initialized with {llm_backend.provider.value} provider")
-else:
-    survey_gen = None
-    survey_trans = None
-    data_analyzer = None
-    print("✗ ConversAI initialization incomplete - no LLM credentials found")
-
-
-# ===========================
-# Survey Generation Functions
-# ===========================
-
-def generate_survey_from_outline(outline: str, survey_type: str, num_questions: int, audience: str):
-    """Generate survey from user outline"""
-    global current_survey
-
-    # Check if backend is initialized
-    if not survey_gen:
-        return (
-            "❌ LLM backend not configured. Please set up API credentials:\n"
-            "- For HuggingFace Spaces: HF_TOKEN is auto-available\n"
-            "- For OpenAI: Set OPENAI_API_KEY\n"
-            "- For Anthropic: Set ANTHROPIC_API_KEY\n"
-            "- For HuggingFace: Set HUGGINGFACE_API_KEY",
-            "",
-            None
-        )
-
-    if not outline or not outline.strip():
-        return "❌ Please provide an outline or topic description.", "", None
-
-    # Validate inputs
-    if num_questions < 1 or num_questions > 50:
-        return "❌ Number of questions must be between 1 and 50.", "", None
-
-    try:
-        # Generate survey
-        survey_data = survey_gen.generate_survey(
-            outline=outline,
-            survey_type=survey_type.lower(),
-            num_questions=num_questions,
-            target_audience=audience
-        )
-
-        current_survey = survey_data
-
-        # Format for display
-        display_text = format_survey_display(survey_data)
-
-        # Save to file for download
-        filepath = save_json_file(survey_data, "survey")
-
-        return (
-            f"✅ Survey generated successfully! Contains {len(survey_data.get('questions', []))} questions.",
-            display_text,
-            filepath
-        )
-
-    except Exception as e:
-        error_msg = f"❌ Error generating survey: {str(e)}"
-        print(f"Survey generation error: {traceback.format_exc()}")
-        return error_msg, "", None
-
-
-def format_survey_display(survey_data: Dict) -> str:
-    """Format survey data for readable display"""
-    output = f"# {survey_data.get('title', 'Survey')}\n\n"
-    output += f"## Introduction\n{survey_data.get('introduction', '')}\n\n"
-    output += "## Questions\n\n"
-
-    for i, q in enumerate(survey_data.get('questions', []), 1):
-        output += f"**{i}. {q.get('question_text', '')}**\n"
-        output += f"   - Type: {q.get('question_type', 'N/A')}\n"
-
-        if q.get('options'):
-            output += "   - Options:\n"
-            for opt in q['options']:
-                output += f"     - {opt}\n"
-
-        if q.get('help_text'):
-            output += f"   - Help: {q['help_text']}\n"
-
-        output += f"   - Required: {'Yes' if q.get('required', False) else 'No'}\n\n"
-
-    output += f"## Closing\n{survey_data.get('closing', '')}\n"
-
-    return output
-
-
-# ===========================
-# Translation Functions
-# ===========================
-
-def translate_current_survey(target_languages: List[str]):
-    """Translate the current survey to selected languages"""
-    global current_survey
-
-    # Check if backend is initialized
-    if not survey_trans:
-        return (
-            "❌ LLM backend not configured. Please set up API credentials in Settings.",
-            "",
-            None
-        )
-
-    if not current_survey:
-        return "❌ Please generate or upload a survey first.", "", None
-
-    if not target_languages:
-        return "❌ Please select at least one target language.", "", None
-
-    try:
-        # Translate to all selected languages
-        translations = {}
-        status_messages = []
-        success_count = 0
-
-        for lang_code in target_languages:
-            try:
-                translated = survey_trans.translate_survey(current_survey, lang_code)
-                translations[lang_code] = translated
-                lang_name = survey_trans._resolve_language(lang_code)
-                status_messages.append(f"✅ Translated to {lang_name}")
-                success_count += 1
-            except Exception as e:
-                lang_name = survey_trans._resolve_language(lang_code)
-                status_messages.append(f"❌ Failed to translate to {lang_name}: {str(e)}")
-                print(f"Translation error for {lang_code}: {traceback.format_exc()}")
-
-        if success_count == 0:
-            return "❌ All translations failed. Please check your LLM configuration.", "", None
-
-        # Format translations for display
-        display_text = ""
-        for lang_code, trans_survey in translations.items():
-            if "error" not in trans_survey:
-                lang_name = survey_trans._resolve_language(lang_code)
-                display_text += f"\n{'='*50}\n"
-                display_text += f"TRANSLATION: {lang_name.upper()}\n"
-                display_text += f"{'='*50}\n\n"
-                display_text += format_survey_display(trans_survey)
-
-        # Save to file for download
-        filepath = save_json_file(translations, "translations")
-
-        status = "\n".join(status_messages)
-        return status, display_text, filepath
-
-    except Exception as e:
-        error_msg = f"❌ Error during translation: {str(e)}"
-        print(f"Translation error: {traceback.format_exc()}")
-        return error_msg, "", None
-
-
-def get_language_choices():
-    """Get language choices for dropdown"""
-    # Get languages directly from SurveyTranslator class (static list)
-    from survey_translator import SurveyTranslator
-    langs = SurveyTranslator.SUPPORTED_LANGUAGES
-    return [f"{code} - {name}" for code, name in langs.items()]
-
-
-# ===========================
-# Data Analysis Functions
-# ===========================
-
-def analyze_survey_data(responses_json: str, questions_json: str = None):
-    """Analyze survey responses"""
-    # Check if backend is initialized
-    if not data_analyzer:
-        return (
-            "❌ LLM backend not configured. Please set up API credentials in Settings.",
-            "",
-            None
-        )
-
-    if not responses_json or not responses_json.strip():
-        return "❌ Please provide survey responses in JSON format.", "", None
-
-    try:
-        # Parse responses
-        responses = json.loads(responses_json)
-        questions = json.loads(questions_json) if questions_json and questions_json.strip() else None
-
-        if not isinstance(responses, list):
-            return "❌ Responses must be a JSON array.", "", None
-
-        if len(responses) == 0:
-            return "❌ No responses to analyze.", "", None
-
-        # Validate questions if provided
-        if questions and not isinstance(questions, list):
-            return "❌ Questions must be a JSON array.", "", None
-
-        # Run analysis
-        analysis_results = data_analyzer.analyze_responses(responses, questions)
-
-        if "error" in analysis_results:
-            return f"❌ Analysis error: {analysis_results['error']}", "", None
-
-        # Generate report
-        report_md = data_analyzer.generate_report(analysis_results, format="markdown")
-
-        # Save both JSON and Markdown
-        json_filepath = save_json_file(analysis_results, "analysis_results")
-        md_filepath = analysis_to_markdown_file(report_md, "analysis_report")
-
-        status_msg = f"✅ Analysis complete! Analyzed {len(responses)} responses."
-        if questions:
-            status_msg += f" Considered {len(questions)} questions."
-
-        return status_msg, report_md, json_filepath
-
-    except json.JSONDecodeError as e:
-        return f"❌ Invalid JSON format: {str(e)}", "", None
-    except Exception as e:
-        error_msg = f"❌ Error during analysis: {str(e)}"
-        print(f"Analysis error: {traceback.format_exc()}")
-        return error_msg, "", None
-
-
-def load_example_responses():
-    """Load example responses for demonstration"""
-    example = [
-        {
-            "q1": "The medication helped reduce my symptoms significantly within the first week.",
-            "q2": "I experienced some mild side effects like drowsiness in the beginning.",
-            "q3": "Overall, I'm satisfied with the treatment and would recommend it to others."
-        },
-        {
-            "q1": "I didn't notice much improvement in my condition after taking the medication.",
-            "q2": "The side effects were quite severe and made it difficult to continue.",
-            "q3": "I had to stop taking it after two weeks due to adverse reactions."
-        },
-        {
-            "q1": "The medication worked well but took about 3-4 weeks to show results.",
-            "q2": "No major side effects, just some occasional nausea.",
-            "q3": "It's been effective for managing my symptoms on a daily basis."
-        }
-    ]
-    return json.dumps(example, indent=2)
-
-
-# ===========================
-# Conversational Research Handlers
-# ===========================
-
-def create_new_flow(flow_name: str, flow_description: str):
-    """Create a new conversation flow"""
-    global current_flow, saved_flows
-
-    if not flow_name or not flow_name.strip():
-        return "❌ Please provide a flow name.", "", None
-
-    try:
-        flow = ConversationFlow(name=flow_name, description=flow_description)
-        current_flow = flow
-        saved_flows[flow.id] = flow
-
-        return (
-            f"✅ Flow '{flow_name}' created successfully!",
-            f"**Flow ID:** {flow.id}\n**Name:** {flow.name}\n**Description:** {flow.description}",
-            flow.id
-        )
-    except Exception as e:
-        return f"❌ Error creating flow: {str(e)}", "", None
-
-
-def load_example_flow():
-    """Load an example conversation flow"""
-    global current_flow, saved_flows
-
-    flow = create_example_flow()
-    current_flow = flow
-    saved_flows[flow.id] = flow
-
-    return (
-        f"✅ Example flow loaded: {flow.name}",
-        display_flow(flow),
-        flow.id
-    )
-
-
-def add_flow_node(flow_id: str, node_content: str, node_type: str):
-    """Add a node to the current flow"""
-    global current_flow, saved_flows
-
-    if not flow_id:
-        return "❌ No flow selected.", ""
-
-    flow = saved_flows.get(flow_id)
-    if not flow:
-        return "❌ Flow not found.", ""
-
-    if not node_content or not node_content.strip():
-        return "❌ Please provide content for the node.", ""
-
-    try:
-        node = ConversationNode(content=node_content, node_type=node_type.lower())
-
-        # Link to previous node if exists
-        if flow.nodes:
-            last_node = flow.nodes[-1]
-            last_node.next = node.id
-
-        flow.add_node(node)
-        current_flow = flow
-
-        return (
-            f"✅ Node added successfully! Total nodes: {len(flow.nodes)}",
-            display_flow(flow)
-        )
-    except Exception as e:
-        return f"❌ Error adding node: {str(e)}", ""
-
-
-def display_flow(flow: ConversationFlow) -> str:
-    """Display flow as markdown"""
-    if not flow or not flow.nodes:
-        return "No flow to display"
-
-    output = f"# {flow.name}\n\n"
-    output += f"**Description:** {flow.description}\n\n"
-    output += f"**Total Steps:** {len(flow.nodes)}\n\n"
-    output += "---\n\n"
-
-    for i, node in enumerate(flow.nodes, 1):
-        output += f"### Step {i}: {node.type.capitalize()}\n\n"
-        output += f"{node.content}\n\n"
-
-    return output
-
-
-def save_current_flow(flow_id: str):
-    """Save the current flow to file"""
-    if not flow_id:
-        return "❌ No flow selected.", None
-
-    flow = saved_flows.get(flow_id)
-    if not flow:
-        return "❌ Flow not found.", None
-
-    try:
-        filepath = save_json_file(flow.to_dict(), "conversation_flow")
-        return f"✅ Flow saved to {filepath}", filepath
-    except Exception as e:
-        return f"❌ Error saving flow: {str(e)}", None
-
-
-def start_conversation_session(flow_id: str):
-    """Start a new conversation session"""
-    global current_session, session_manager
-
-    if not flow_id:
-        return [], "❌ Please select a flow first."
-
-    flow = saved_flows.get(flow_id)
-    if not flow:
-        return [], "❌ Flow not found."
-
-    if not llm_backend:
-        return [], "❌ LLM backend not initialized."
-
-    try:
-        # Create session
-        session = session_manager.create_session(flow_id=flow.id, flow_name=flow.name)
-        current_session = session
-
-        # Create moderator
-        moderator = ConversationModerator(llm_backend, flow)
-
-        # Start conversation
-        opening_message = moderator.start_conversation(session)
-
-        # Return chat history in Gradio format
-        return [[None, opening_message]], f"✅ Conversation started! Session ID: {session.id}"
-
-    except Exception as e:
-        return [], f"❌ Error starting conversation: {str(e)}"
-
-
-def chat_with_moderator(user_message: str, history: List):
-    """Handle chat messages with the AI moderator"""
-    global current_session
-
-    if not current_session:
-        return history, "❌ No active session. Please start a conversation first."
-
-    if not llm_backend:
-        return history, "❌ LLM backend not initialized."
-
-    if not user_message or not user_message.strip():
-        return history, "❌ Please enter a message."
-
-    try:
-        # Get the flow
-        flow = saved_flows.get(current_session.flow_id)
-        if not flow:
-            return history, "❌ Flow not found."
-
-        # Create moderator
-        moderator = ConversationModerator(llm_backend, flow)
-
-        # Process user response
-        ai_response = moderator.process_user_response(current_session, user_message)
-
-        # Update history
-        history.append([user_message, ai_response])
-
-        status = f"Session: {current_session.id} | Turns: {current_session.get_turn_count()}"
-        if current_session.status == "completed":
-            status += " | ✅ Conversation completed"
-
-        return history, status
-
-    except Exception as e:
-        return history, f"❌ Error: {str(e)}"
-
-
-def export_conversation():
-    """Export the current conversation"""
-    global current_session
-
-    if not current_session:
-        return "❌ No active session to export.", None
-
-    try:
-        filepath = conversation_to_transcript(current_session)
-        return f"✅ Conversation exported to {filepath}", filepath
-    except Exception as e:
-        return f"❌ Error exporting conversation: {str(e)}", None
-
-
-# ===========================
-# Gradio Interface
-# ===========================
-
-def create_interface():
-    """Create the main Gradio interface"""
-
-    with gr.Blocks(
-        title="ProjectEcho - Qualitative Research Assistant",
-        theme=gr.themes.Soft(primary_hue="blue", secondary_hue="slate")
-    ) as app:
-
-        gr.Markdown("""
-        # ProjectEcho - Your AI-Powered Qualitative Research Assistant
-
-        Battle the blank page, reach global audiences, and uncover insights with AI assistance.
-        """)
-
-        # Show backend status
-        if llm_backend:
-            status_msg = f"✅ **Active LLM Provider:** {llm_backend.provider.value.upper()} | Model: {llm_backend.model}"
-            bg_color = "rgba(0, 255, 0, 0.1)"
-        else:
-            status_msg = """⚠️ **LLM Provider Not Configured**
-
-**To use this app, you need to configure an LLM provider:**
-
-1. **Easiest (HuggingFace Spaces):** Make sure your Space is PUBLIC and HF_TOKEN will be auto-available
-2. **Best Quality:** Add `OPENAI_API_KEY` in Space Settings → Variables
-3. **Alternative:** Add `ANTHROPIC_API_KEY` or `HUGGINGFACE_API_KEY`
-
-See the **About** tab for detailed instructions."""
-            bg_color = "rgba(255, 165, 0, 0.2)"
-
-        gr.Markdown(f'<div style="background-color: {bg_color}; padding: 15px; border-radius: 5px; margin: 10px 0; border-left: 4px solid #FF6B6B;">{status_msg}</div>')
-
-        with gr.Tabs() as tabs:
-
-            # ========== SURVEY GENERATION TAB ==========
-            with gr.Tab("📝 Generate Survey"):
-                gr.Markdown("""
-                ## Battle the Blank Page
-                Share an outline and get AI-powered surveys drafted in minutes,
-                complete with industry best practices.
-                """)
-
-                with gr.Row():
-                    with gr.Column(scale=1):
-                        outline_input = gr.Textbox(
-                            label="Your Survey Outline or Topic",
-                            placeholder="Example: I want to understand patient experiences with a new diabetes medication, focusing on effectiveness, side effects, and quality of life impacts.",
-                            lines=6
-                        )
-
-                        survey_type_input = gr.Radio(
-                            label="Survey Type",
-                            choices=["Qualitative", "Quantitative", "Mixed"],
-                            value="Qualitative"
-                        )
-
-                        num_questions_input = gr.Slider(
-                            label="Number of Questions",
-                            minimum=5,
-                            maximum=25,
-                            value=10,
-                            step=1
-                        )
-
-                        audience_input = gr.Textbox(
-                            label="Target Audience",
-                            placeholder="Example: Adults aged 30-65 with Type 2 diabetes",
-                            value="General audience"
-                        )
-
-                        generate_btn = gr.Button("🚀 Generate Survey", variant="primary", size="lg")
-
-                    with gr.Column(scale=1):
-                        gen_status = gr.Textbox(label="Status", interactive=False)
-                        gen_output = gr.Markdown(label="Generated Survey")
-
-                gen_download = gr.File(label="Download Survey JSON", visible=False)
-
-                # Event handlers
-                generate_btn.click(
-                    fn=generate_survey_from_outline,
-                    inputs=[outline_input, survey_type_input, num_questions_input, audience_input],
-                    outputs=[gen_status, gen_output, gen_download]
-                ).then(
-                    fn=lambda x: gr.File(value=x, visible=True) if x else gr.File(visible=False),
-                    inputs=[gen_download],
-                    outputs=[gen_download]
-                )
-
-            # ========== TRANSLATION TAB ==========
-            with gr.Tab("🌍 Translate Survey"):
-                gr.Markdown("""
-                ## Reach Global Audiences
-                Translate your surveys automatically to streamline efforts and reach wider audiences.
-                """)
-
-                with gr.Row():
-                    with gr.Column(scale=1):
-                        gr.Markdown("### Select Target Languages")
-
-                        # Create checkboxes for popular languages
-                        lang_checkboxes = gr.CheckboxGroup(
-                            label="Languages",
-                            choices=get_language_choices(),
-                            value=[]
-                        )
-
-                        translate_btn = gr.Button("🌐 Translate Survey", variant="primary", size="lg")
-
-                        gr.Markdown("""
-                        **Note:** Make sure you've generated a survey first, or upload one using the JSON format.
-                        """)
-
-                    with gr.Column(scale=1):
-                        trans_status = gr.Textbox(label="Translation Status", interactive=False)
-                        trans_output = gr.Markdown(label="Translations")
-
-                trans_download = gr.File(label="Download Translations JSON", visible=False)
-
-                # Event handlers
-                def extract_lang_codes(selected_items):
-                    """Extract language codes from checkbox selections"""
-                    return [item.split(" - ")[0] for item in selected_items]
-
-                translate_btn.click(
-                    fn=lambda x: translate_current_survey(extract_lang_codes(x)),
-                    inputs=[lang_checkboxes],
-                    outputs=[trans_status, trans_output, trans_download]
-                ).then(
-                    fn=lambda x: gr.File(value=x, visible=True) if x else gr.File(visible=False),
-                    inputs=[trans_download],
-                    outputs=[trans_download]
-                )
-
-            # ========== ANALYSIS TAB ==========
-            with gr.Tab("📊 Analyze Data"):
-                gr.Markdown("""
-                ## Uncover Key Insights
-                Upload your survey responses and get AI-assisted summaries of key findings,
-                themes, and trends.
-                """)
-
-                with gr.Row():
-                    with gr.Column(scale=1):
-                        responses_input = gr.Textbox(
-                            label="Survey Responses (JSON)",
-                            placeholder='[{"q1": "response 1", "q2": "response 2"}, ...]',
-                            lines=10
-                        )
-
-                        questions_input = gr.Textbox(
-                            label="Questions (JSON, Optional)",
-                            placeholder='[{"question_text": "What is your experience?", ...}]',
-                            lines=5
-                        )
-
-                        with gr.Row():
-                            analyze_btn = gr.Button("🔍 Analyze Data", variant="primary", size="lg")
-                            example_btn = gr.Button("Load Example", variant="secondary")
-
-                    with gr.Column(scale=1):
-                        analysis_status = gr.Textbox(label="Status", interactive=False)
-                        analysis_output = gr.Markdown(label="Analysis Report")
-
-                analysis_download = gr.File(label="Download Analysis JSON", visible=False)
-
-                # Event handlers
-                analyze_btn.click(
-                    fn=analyze_survey_data,
-                    inputs=[responses_input, questions_input],
-                    outputs=[analysis_status, analysis_output, analysis_download]
-                ).then(
-                    fn=lambda x: gr.File(value=x, visible=True) if x else gr.File(visible=False),
-                    inputs=[analysis_download],
-                    outputs=[analysis_download]
-                )
-
-                example_btn.click(
-                    fn=load_example_responses,
-                    outputs=[responses_input]
-                )
-
-            # ========== CONVERSATIONAL RESEARCH TAB ==========
-            with gr.Tab("💬 Conversational Research"):
-                gr.Markdown("""
-                ## AI-Moderated Conversations
-                Design conversation flows and conduct AI-powered qualitative interviews with respondents.
-                """)
-
-                with gr.Tabs():
-                    # Design Flow Sub-Tab
-                    with gr.Tab("🎨 Design Flow"):
-                        gr.Markdown("""
-                        ### Create Conversation Flows
-                        Design custom conversation paths for AI-moderated interviews.
-                        """)
-
-                        with gr.Row():
-                            with gr.Column(scale=1):
-                                gr.Markdown("#### Flow Setup")
-
-                                flow_name_input = gr.Textbox(
-                                    label="Flow Name",
-                                    placeholder="e.g., HCP Interview for New Dermatology Product",
-                                    value=""
-                                )
-
-                                flow_desc_input = gr.Textbox(
-                                    label="Flow Description",
-                                    placeholder="Describe the purpose of this conversation flow...",
-                                    lines=3
-                                )
-
-                                with gr.Row():
-                                    create_flow_btn = gr.Button("✨ Create New Flow", variant="primary")
-                                    load_example_flow_btn = gr.Button("📋 Load Example", variant="secondary")
-
-                                flow_id_state = gr.State(value="")
-
-                                gr.Markdown("#### Add Steps to Flow")
-
-                                node_content_input = gr.Textbox(
-                                    label="Question/Message",
-                                    placeholder="Enter the question or message for this step...",
-                                    lines=4
-                                )
-
-                                node_type_input = gr.Radio(
-                                    label="Step Type",
-                                    choices=["Question", "End"],
-                                    value="Question"
-                                )
-
-                                add_node_btn = gr.Button("➕ Add Step", variant="secondary")
-
-                                save_flow_btn = gr.Button("💾 Save Flow", variant="primary")
-
-                            with gr.Column(scale=1):
-                                flow_status = gr.Textbox(label="Status", interactive=False)
-                                flow_display = gr.Markdown(label="Flow Preview", value="No flow created yet")
-
-                        flow_download = gr.File(label="Download Flow JSON", visible=False)
-
-                        # Event handlers for flow design
-                        create_flow_btn.click(
-                            fn=create_new_flow,
-                            inputs=[flow_name_input, flow_desc_input],
-                            outputs=[flow_status, flow_display, flow_id_state]
-                        )
-
-                        load_example_flow_btn.click(
-                            fn=load_example_flow,
-                            outputs=[flow_status, flow_display, flow_id_state]
-                        )
-
-                        add_node_btn.click(
-                            fn=add_flow_node,
-                            inputs=[flow_id_state, node_content_input, node_type_input],
-                            outputs=[flow_status, flow_display]
-                        ).then(
-                            fn=lambda: "",
-                            outputs=[node_content_input]
-                        )
-
-                        save_flow_btn.click(
-                            fn=save_current_flow,
-                            inputs=[flow_id_state],
-                            outputs=[flow_status, flow_download]
-                        ).then(
-                            fn=lambda x: gr.File(value=x, visible=True) if x else gr.File(visible=False),
-                            inputs=[flow_download],
-                            outputs=[flow_download]
-                        )
-
-                    # Conduct Interview Sub-Tab
-                    with gr.Tab("🎙️ Conduct Interview"):
-                        gr.Markdown("""
-                        ### AI-Moderated Interview
-                        Start a conversation session with the AI moderator using your designed flow.
-                        """)
-
-                        with gr.Row():
-                            with gr.Column(scale=1):
-                                conversation_flow_selector = gr.State(value="")
-
-                                gr.Markdown("""
-                                **Instructions:**
-                                1. Design a flow in the 'Design Flow' tab first (or load the example)
-                                2. Click 'Start Conversation' to begin
-                                3. The AI moderator will ask questions from your flow
-                                4. The AI adapts with follow-up questions based on responses
-                                5. Export the conversation when finished
-                                """)
-
-                                with gr.Row():
-                                    start_conversation_btn = gr.Button("🚀 Start Conversation", variant="primary")
-                                    export_conversation_btn = gr.Button("📥 Export Conversation", variant="secondary")
-
-                                conversation_status = gr.Textbox(label="Session Status", interactive=False)
-                                conversation_download = gr.File(label="Download Transcript", visible=False)
-
-                            with gr.Column(scale=1):
-                                chatbot = gr.Chatbot(
-                                    label="AI-Moderated Interview",
-                                    height=500
-                                )
-
-                                msg_input = gr.Textbox(
-                                    label="Your Response",
-                                    placeholder="Type your response here...",
-                                    lines=2
-                                )
-
-                                with gr.Row():
-                                    submit_btn = gr.Button("Send", variant="primary")
-                                    clear_btn = gr.Button("Clear")
-
-                        # Chat event handlers
-                        def user_submit(user_message, history):
-                            """Handle user message submission"""
-                            if not user_message:
-                                return history, history, ""
-                            return history, history + [[user_message, None]], ""
-
-                        def bot_respond(history):
-                            """Get bot response"""
-                            if not history or history[-1][1] is not None:
-                                return history, ""
-
-                            user_msg = history[-1][0]
-                            updated_history, status = chat_with_moderator(user_msg, history[:-1])
-                            return updated_history, status
-
-                        # Start conversation
-                        start_conversation_btn.click(
-                            fn=lambda: saved_flows[list(saved_flows.keys())[-1]].id if saved_flows else "",
-                            outputs=[conversation_flow_selector]
-                        ).then(
-                            fn=start_conversation_session,
-                            inputs=[conversation_flow_selector],
-                            outputs=[chatbot, conversation_status]
-                        )
-
-                        # Message submission
-                        msg_input.submit(
-                            fn=user_submit,
-                            inputs=[msg_input, chatbot],
-                            outputs=[chatbot, chatbot, msg_input],
-                            queue=False
-                        ).then(
-                            fn=bot_respond,
-                            inputs=[chatbot],
-                            outputs=[chatbot, conversation_status]
-                        )
-
-                        submit_btn.click(
-                            fn=user_submit,
-                            inputs=[msg_input, chatbot],
-                            outputs=[chatbot, chatbot, msg_input],
-                            queue=False
-                        ).then(
-                            fn=bot_respond,
-                            inputs=[chatbot],
-                            outputs=[chatbot, conversation_status]
-                        )
-
-                        clear_btn.click(lambda: None, None, chatbot, queue=False)
-
-                        # Export conversation
-                        export_conversation_btn.click(
-                            fn=export_conversation,
-                            outputs=[conversation_status, conversation_download]
-                        ).then(
-                            fn=lambda x: gr.File(value=x, visible=True) if x else gr.File(visible=False),
-                            inputs=[conversation_download],
-                            outputs=[conversation_download]
-                        )
-
-            # ========== ABOUT TAB ==========
-            with gr.Tab("ℹ️ About"):
-                gr.Markdown("""
-                ## About ConversAI
-
-                ConversAI is a comprehensive qualitative research assistant that helps you:
-
-                ### 🎯 Generate Surveys
-                - Create professional surveys from simple outlines
-                - Follow industry best practices automatically
-                - Save hours of questionnaire design time
-
-                ### 🌍 Translate Globally
-                - Reach audiences in 18+ languages
-                - Maintain cultural appropriateness
-                - Expand your research scope effortlessly
-
-                ### 📊 Analyze Results
-                - Extract key themes automatically
-                - Identify patterns and trends
-                - Generate actionable insights
-
-                ### 🔧 Configuration Guide
-
-                **For HuggingFace Spaces (Recommended):**
-
-                No configuration needed! The app automatically uses the HF Inference API with the built-in `HF_TOKEN`.
-
-                **Supported Models:**
-                - Default: `mistralai/Mixtral-8x7B-Instruct-v0.1`
-                - You can change by setting `LLM_MODEL` environment variable
-
-                **For Other LLM Providers:**
-
-                Add these environment variables in your Space Settings:
-
-                1. **OpenAI** (Best quality, paid):
-                   - `LLM_PROVIDER=openai`
-                   - `OPENAI_API_KEY=sk-your-key`
-
-                2. **Anthropic Claude** (Best reasoning, paid):
-                   - `LLM_PROVIDER=anthropic`
-                   - `ANTHROPIC_API_KEY=your-key`
-
-                3. **Custom HuggingFace Model**:
-                   - `LLM_PROVIDER=huggingface`
-                   - `LLM_MODEL=your-model-name`
-
-                **💡 Pro Tip:** For production use, we recommend OpenAI or Anthropic for faster, more reliable results.
-
-                **Supported LLM Providers:**
-                - HuggingFace Inference API (Free tier available)
-                - OpenAI (GPT-4, GPT-4o-mini, GPT-3.5)
-                - Anthropic (Claude 3.5 Sonnet, Claude 3 Opus)
-                - LM Studio (local development only)
-
-                ### 📄 Data Privacy
-
-                - All processing is done through your configured LLM provider
-                - No data is stored permanently by this application
-                - Survey data and responses remain in your control
-
-                ### 🚀 Getting Started
-
-                1. **Generate** a survey from your research outline
-                2. **Translate** it to reach global audiences
-                3. Collect responses from participants
-                4. **Analyze** the data to uncover insights
-
-                ---
-
-                Built with ❤️ using Gradio and state-of-the-art LLMs
-                """)
-
-        return app
-
-
-# ===========================
-# Main Entry Point
-# ===========================
-
-if __name__ == "__main__":
-    demo = create_interface()
-
-    # Launch with appropriate settings
-    demo.launch(
-        server_name="0.0.0.0",  # Allow external access
-        server_port=7860,  # Standard HF Spaces port
-        share=False,  # Don't create a public link (HF Spaces handles this)
-        show_error=True
-    )
+"""
+Project Echo - AI-Powered Qualitative Research Assistant
+Production-grade survey generation, translation, and analysis platform
+"""
+import gradio as gr
+import json
+import os
+import traceback
+from typing import Dict, List, Optional
+
+from llm_backend import LLMBackend, LLMProvider
+from survey_generator import SurveyGenerator
+from survey_translator import SurveyTranslator
+from data_analyzer import DataAnalyzer
+from export_utils import (save_json_file, survey_to_csv, analysis_to_markdown_file,
+                          conversation_to_transcript, conversation_to_json, conversation_to_csv,
+                          flow_to_markdown)
+from conversation_flow import ConversationFlow, ConversationNode, create_example_flow
+from conversation_session import ConversationSession, SessionManager
+from conversation_moderator import ConversationModerator
+from conversation_analytics import ConversationAnalytics
+
+
+# Global state for current survey
+current_survey = None
+current_responses = []
+
+# Global state for conversational research
+current_flow = None
+session_manager = SessionManager()
+current_session = None
+saved_flows = {}
+
+
+def initialize_backend():
+    """Initialize LLM backend based on environment"""
+    try:
+        # Debug: Print all environment variables related to LLM
+        print("=== LLM Backend Initialization ===")
+        print(f"HF_TOKEN: {'SET' if os.getenv('HF_TOKEN') else 'NOT SET'}")
+        print(f"HUGGINGFACE_API_KEY: {'SET' if os.getenv('HUGGINGFACE_API_KEY') else 'NOT SET'}")
+        print(f"OPENAI_API_KEY: {'SET' if os.getenv('OPENAI_API_KEY') else 'NOT SET'}")
+        print(f"ANTHROPIC_API_KEY: {'SET' if os.getenv('ANTHROPIC_API_KEY') else 'NOT SET'}")
+        print(f"LLM_PROVIDER: {os.getenv('LLM_PROVIDER', 'NOT SET')}")
+
+        # Check for explicit provider setting
+        provider_env = os.getenv("LLM_PROVIDER", "").lower()
+
+        # Priority 1: Explicitly set provider
+        if provider_env == "openai" and os.getenv("OPENAI_API_KEY"):
+            print("Using OpenAI (explicit)")
+            return LLMBackend(provider=LLMProvider.OPENAI)
+        elif provider_env == "anthropic" and os.getenv("ANTHROPIC_API_KEY"):
+            print("Using Anthropic (explicit)")
+            return LLMBackend(provider=LLMProvider.ANTHROPIC)
+        elif provider_env == "huggingface" and (os.getenv("HUGGINGFACE_API_KEY") or os.getenv("HF_TOKEN")):
+            api_key = os.getenv("HUGGINGFACE_API_KEY") or os.getenv("HF_TOKEN")
+            print("Using HuggingFace (explicit)")
+            return LLMBackend(provider=LLMProvider.HUGGINGFACE, api_key=api_key)
+        elif provider_env == "lm_studio":
+            print("Using LM Studio (explicit)")
+            return LLMBackend(provider=LLMProvider.LM_STUDIO)
+
+        # Priority 2: Auto-detect based on available credentials
+        # HF_TOKEN is automatically available in HF Spaces, so check it first
+        hf_token = os.getenv("HF_TOKEN") or os.getenv("HUGGINGFACE_API_KEY")
+        if hf_token:
+            print(f"Auto-detected HuggingFace credentials, using HF Inference API")
+            print(f"Token preview: {hf_token[:10]}...")
+            return LLMBackend(provider=LLMProvider.HUGGINGFACE, api_key=hf_token)
+        elif os.getenv("OPENAI_API_KEY"):
+            print(f"Auto-detected OpenAI credentials")
+            return LLMBackend(provider=LLMProvider.OPENAI)
+        elif os.getenv("ANTHROPIC_API_KEY"):
+            print(f"Auto-detected Anthropic credentials")
+            return LLMBackend(provider=LLMProvider.ANTHROPIC)
+        else:
+            # No credentials found - return None to show error in UI
+            print("="*60)
+            print("WARNING: No LLM provider credentials found!")
+            print("="*60)
+            print("For HuggingFace Spaces:")
+            print("  - HF_TOKEN should be automatically available")
+            print("  - Make sure your Space is PUBLIC")
+            print("  - Or add HUGGINGFACE_API_KEY in Settings")
+            print("")
+            print("For other providers, set one of:")
+            print("  - OPENAI_API_KEY")
+            print("  - ANTHROPIC_API_KEY")
+            print("  - HUGGINGFACE_API_KEY")
+            print("="*60)
+            return None
+
+    except Exception as e:
+        print(f"Error during backend initialization: {e}")
+        import traceback
+        traceback.print_exc()
+        return None
+
+
+# Initialize components
+llm_backend = initialize_backend()
+
+# Only initialize if backend is available
+if llm_backend:
+    survey_gen = SurveyGenerator(llm_backend)
+    survey_trans = SurveyTranslator(llm_backend)
+    data_analyzer = DataAnalyzer(llm_backend)
+    print(f"✓ Project Echo initialized with {llm_backend.provider.value} provider")
+else:
+    survey_gen = None
+    survey_trans = None
+    data_analyzer = None
+    print("✗ Project Echo initialization incomplete - no LLM credentials found")
+
+
+# ===========================
+# Survey Generation Functions
+# ===========================
+
+def generate_survey_from_outline(outline: str, survey_type: str, num_questions: int, audience: str):
+    """Generate survey from user outline"""
+    global current_survey
+
+    # Check if backend is initialized
+    if not survey_gen:
+        return (
+            "❌ LLM backend not configured. Please set up API credentials:\n"
+            "- For HuggingFace Spaces: HF_TOKEN is auto-available\n"
+            "- For OpenAI: Set OPENAI_API_KEY\n"
+            "- For Anthropic: Set ANTHROPIC_API_KEY\n"
+            "- For HuggingFace: Set HUGGINGFACE_API_KEY",
+            "",
+            None
+        )
+
+    if not outline or not outline.strip():
+        return "❌ Please provide an outline or topic description.", "", None
+
+    # Validate inputs
+    if num_questions < 1 or num_questions > 50:
+        return "❌ Number of questions must be between 1 and 50.", "", None
+
+    try:
+        # Generate survey
+        survey_data = survey_gen.generate_survey(
+            outline=outline,
+            survey_type=survey_type.lower(),
+            num_questions=num_questions,
+            target_audience=audience
+        )
+
+        current_survey = survey_data
+
+        # Format for display
+        display_text = format_survey_display(survey_data)
+
+        # Save to file for download
+        filepath = save_json_file(survey_data, "survey")
+
+        return (
+            f"✅ Survey generated successfully! Contains {len(survey_data.get('questions', []))} questions.",
+            display_text,
+            filepath
+        )
+
+    except Exception as e:
+        error_msg = f"❌ Error generating survey: {str(e)}"
+        print(f"Survey generation error: {traceback.format_exc()}")
+        return error_msg, "", None
+
+
+def format_survey_display(survey_data: Dict) -> str:
+    """Format survey data for readable display"""
+    output = f"# {survey_data.get('title', 'Survey')}\n\n"
+    output += f"## Introduction\n{survey_data.get('introduction', '')}\n\n"
+    output += "## Questions\n\n"
+
+    for i, q in enumerate(survey_data.get('questions', []), 1):
+        output += f"**{i}. {q.get('question_text', '')}**\n"
+        output += f"   - Type: {q.get('question_type', 'N/A')}\n"
+
+        if q.get('options'):
+            output += "   - Options:\n"
+            for opt in q['options']:
+                output += f"     - {opt}\n"
+
+        if q.get('help_text'):
+            output += f"   - Help: {q['help_text']}\n"
+
+        output += f"   - Required: {'Yes' if q.get('required', False) else 'No'}\n\n"
+
+    output += f"## Closing\n{survey_data.get('closing', '')}\n"
+
+    return output
+
+
+# ===========================
+# Translation Functions
+# ===========================
+
+def translate_current_survey(target_languages: List[str]):
+    """Translate the current survey to selected languages"""
+    global current_survey
+
+    # Check if backend is initialized
+    if not survey_trans:
+        return (
+            "❌ LLM backend not configured. Please set up API credentials in Settings.",
+            "",
+            None
+        )
+
+    if not current_survey:
+        return "❌ Please generate or upload a survey first.", "", None
+
+    if not target_languages:
+        return "❌ Please select at least one target language.", "", None
+
+    try:
+        # Translate to all selected languages
+        translations = {}
+        status_messages = []
+        success_count = 0
+
+        for lang_code in target_languages:
+            try:
+                translated = survey_trans.translate_survey(current_survey, lang_code)
+                translations[lang_code] = translated
+                lang_name = survey_trans._resolve_language(lang_code)
+                status_messages.append(f"✅ Translated to {lang_name}")
+                success_count += 1
+            except Exception as e:
+                lang_name = survey_trans._resolve_language(lang_code)
+                status_messages.append(f"❌ Failed to translate to {lang_name}: {str(e)}")
+                print(f"Translation error for {lang_code}: {traceback.format_exc()}")
+
+        if success_count == 0:
+            return "❌ All translations failed. Please check your LLM configuration.", "", None
+
+        # Format translations for display
+        display_text = ""
+        for lang_code, trans_survey in translations.items():
+            if "error" not in trans_survey:
+                lang_name = survey_trans._resolve_language(lang_code)
+                display_text += f"\n{'='*50}\n"
+                display_text += f"TRANSLATION: {lang_name.upper()}\n"
+                display_text += f"{'='*50}\n\n"
+                display_text += format_survey_display(trans_survey)
+
+        # Save to file for download
+        filepath = save_json_file(translations, "translations")
+
+        status = "\n".join(status_messages)
+        return status, display_text, filepath
+
+    except Exception as e:
+        error_msg = f"❌ Error during translation: {str(e)}"
+        print(f"Translation error: {traceback.format_exc()}")
+        return error_msg, "", None
+
+
+def get_language_choices():
+    """Get language choices for dropdown"""
+    # Get languages directly from SurveyTranslator class (static list)
+    from survey_translator import SurveyTranslator
+    langs = SurveyTranslator.SUPPORTED_LANGUAGES
+    return [f"{code} - {name}" for code, name in langs.items()]
+
+
+# ===========================
+# Data Analysis Functions
+# ===========================
+
+def analyze_survey_data(responses_json: str, questions_json: str = None):
+    """Analyze survey responses"""
+    # Check if backend is initialized
+    if not data_analyzer:
+        return (
+            "❌ LLM backend not configured. Please set up API credentials in Settings.",
+            "",
+            None
+        )
+
+    if not responses_json or not responses_json.strip():
+        return "❌ Please provide survey responses in JSON format.", "", None
+
+    try:
+        # Parse responses
+        responses = json.loads(responses_json)
+        questions = json.loads(questions_json) if questions_json and questions_json.strip() else None
+
+        if not isinstance(responses, list):
+            return "❌ Responses must be a JSON array.", "", None
+
+        if len(responses) == 0:
+            return "❌ No responses to analyze.", "", None
+
+        # Validate questions if provided
+        if questions and not isinstance(questions, list):
+            return "❌ Questions must be a JSON array.", "", None
+
+        # Run analysis
+        analysis_results = data_analyzer.analyze_responses(responses, questions)
+
+        if "error" in analysis_results:
+            return f"❌ Analysis error: {analysis_results['error']}", "", None
+
+        # Generate report
+        report_md = data_analyzer.generate_report(analysis_results, format="markdown")
+
+        # Save both JSON and Markdown
+        json_filepath = save_json_file(analysis_results, "analysis_results")
+        md_filepath = analysis_to_markdown_file(report_md, "analysis_report")
+
+        status_msg = f"✅ Analysis complete! Analyzed {len(responses)} responses."
+        if questions:
+            status_msg += f" Considered {len(questions)} questions."
+
+        return status_msg, report_md, json_filepath
+
+    except json.JSONDecodeError as e:
+        return f"❌ Invalid JSON format: {str(e)}", "", None
+    except Exception as e:
+        error_msg = f"❌ Error during analysis: {str(e)}"
+        print(f"Analysis error: {traceback.format_exc()}")
+        return error_msg, "", None
+
+
+def load_example_responses():
+    """Load example responses for demonstration"""
+    example = [
+        {
+            "q1": "The medication helped reduce my symptoms significantly within the first week.",
+            "q2": "I experienced some mild side effects like drowsiness in the beginning.",
+            "q3": "Overall, I'm satisfied with the treatment and would recommend it to others."
+        },
+        {
+            "q1": "I didn't notice much improvement in my condition after taking the medication.",
+            "q2": "The side effects were quite severe and made it difficult to continue.",
+            "q3": "I had to stop taking it after two weeks due to adverse reactions."
+        },
+        {
+            "q1": "The medication worked well but took about 3-4 weeks to show results.",
+            "q2": "No major side effects, just some occasional nausea.",
+            "q3": "It's been effective for managing my symptoms on a daily basis."
+        }
+    ]
+    return json.dumps(example, indent=2)
+
+
+# ===========================
+# Conversational Research Handlers
+# ===========================
+
+def create_new_flow(flow_name: str, flow_description: str):
+    """Create a new conversation flow with AI-generated initial structure"""
+    global current_flow, saved_flows, llm_backend
+
+    if not flow_name or not flow_name.strip():
+        return "❌ Please provide a flow name.", "", None
+
+    if not flow_description or not flow_description.strip():
+        return "❌ Please provide a description of what you want to discuss in this flow.", "", None
+
+    if not llm_backend:
+        return "❌ LLM backend not configured. Cannot generate flow.", "", None
+
+    try:
+        # Create empty flow
+        flow = ConversationFlow(name=flow_name, description=flow_description)
+
+        # Generate initial conversation structure using AI
+        success, message = flow.generate_flow_with_ai(llm_backend, num_questions=5)
+
+        if not success:
+            return f"⚠️ Flow created but generation failed: {message}", display_flow(flow), None
+
+        current_flow = flow
+        saved_flows[flow.id] = flow
+
+        status_msg = f"✅ Flow '{flow_name}' created with {len(flow.nodes)} conversation steps!"
+
+        return (
+            status_msg,
+            display_flow(flow),
+            flow.id
+        )
+    except Exception as e:
+        error_msg = f"❌ Error creating flow: {str(e)}"
+        print(f"Flow creation error: {traceback.format_exc()}")
+        return error_msg, "", None
+
+
+def regenerate_flow_content(flow_id: str):
+    """Regenerate the conversation flow nodes using AI"""
+    global saved_flows, current_flow, llm_backend
+
+    if not flow_id:
+        return "❌ No flow selected.", ""
+
+    flow = saved_flows.get(flow_id)
+    if not flow:
+        return "❌ Flow not found.", ""
+
+    if not llm_backend:
+        return "❌ LLM backend not configured.", ""
+
+    try:
+        # Clear existing nodes
+        flow.nodes = []
+
+        # Regenerate with AI
+        success, message = flow.generate_flow_with_ai(llm_backend, num_questions=5)
+
+        if not success:
+            return f"⚠️ Regeneration failed: {message}", ""
+
+        current_flow = flow
+
+        return (
+            f"✅ Flow regenerated with {len(flow.nodes)} new steps!",
+            display_flow(flow)
+        )
+    except Exception as e:
+        return f"❌ Error regenerating flow: {str(e)}", ""
+
+
+def load_example_flow():
+    """Load an example conversation flow"""
+    global current_flow, saved_flows
+
+    flow = create_example_flow()
+    current_flow = flow
+    saved_flows[flow.id] = flow
+
+    return (
+        f"✅ Example flow loaded: {flow.name}",
+        display_flow(flow),
+        flow.id
+    )
+
+
+def add_flow_node(flow_id: str, node_content: str, node_type: str):
+    """Add a node to the current flow"""
+    global current_flow, saved_flows
+
+    if not flow_id:
+        return "❌ No flow selected.", ""
+
+    flow = saved_flows.get(flow_id)
+    if not flow:
+        return "❌ Flow not found.", ""
+
+    if not node_content or not node_content.strip():
+        return "❌ Please provide content for the node.", ""
+
+    try:
+        node = ConversationNode(content=node_content, node_type=node_type.lower())
+
+        # Link to previous node if exists
+        if flow.nodes:
+            last_node = flow.nodes[-1]
+            last_node.next = node.id
+
+        flow.add_node(node)
+        current_flow = flow
+
+        return (
+            f"✅ Node added successfully! Total nodes: {len(flow.nodes)}",
+            display_flow(flow)
+        )
+    except Exception as e:
+        return f"❌ Error adding node: {str(e)}", ""
+
+
+def display_flow(flow: ConversationFlow) -> str:
+    """Display flow as markdown"""
+    if not flow or not flow.nodes:
+        return "No flow to display"
+
+    output = f"# {flow.name}\n\n"
+    output += f"**Description:** {flow.description}\n\n"
+    output += f"**Total Steps:** {len(flow.nodes)}\n\n"
+    output += "---\n\n"
+
+    for i, node in enumerate(flow.nodes, 1):
+        output += f"### Step {i}: {node.type.capitalize()}\n\n"
+        output += f"{node.content}\n\n"
+
+    return output
+
+
+def save_current_flow(flow_id: str):
+    """Save the current flow to file"""
+    if not flow_id:
+        return "❌ No flow selected.", None
+
+    flow = saved_flows.get(flow_id)
+    if not flow:
+        return "❌ Flow not found.", None
+
+    try:
+        filepath = save_json_file(flow.to_dict(), "conversation_flow")
+        return f"✅ Flow saved to {filepath}", filepath
+    except Exception as e:
+        return f"❌ Error saving flow: {str(e)}", None
+
+
+def start_conversation_session(flow_id: str):
+    """Start a new conversation session"""
+    global current_session, session_manager
+
+    if not flow_id:
+        return [], "❌ Please select a flow first."
+
+    flow = saved_flows.get(flow_id)
+    if not flow:
+        return [], "❌ Flow not found."
+
+    if not llm_backend:
+        return [], "❌ LLM backend not initialized."
+
+    try:
+        # Create session
+        session = session_manager.create_session(flow_id=flow.id, flow_name=flow.name)
+        current_session = session
+
+        # Create moderator
+        moderator = ConversationModerator(llm_backend, flow)
+
+        # Start conversation
+        opening_message = moderator.start_conversation(session)
+
+        # Return chat history in Gradio format
+        return [[None, opening_message]], f"✅ Conversation started! Session ID: {session.id}"
+
+    except Exception as e:
+        return [], f"❌ Error starting conversation: {str(e)}"
+
+
+def chat_with_moderator(user_message: str, history: List):
+    """Handle chat messages with the AI moderator"""
+    global current_session
+
+    if not current_session:
+        return history, "❌ No active session. Please start a conversation first."
+
+    if not llm_backend:
+        return history, "❌ LLM backend not initialized."
+
+    if not user_message or not user_message.strip():
+        return history, "❌ Please enter a message."
+
+    try:
+        # Get the flow
+        flow = saved_flows.get(current_session.flow_id)
+        if not flow:
+            return history, "❌ Flow not found."
+
+        # Create moderator
+        moderator = ConversationModerator(llm_backend, flow)
+
+        # Process user response
+        ai_response = moderator.process_user_response(current_session, user_message)
+
+        # Update history
+        history.append([user_message, ai_response])
+
+        status = f"Session: {current_session.id} | Turns: {current_session.get_turn_count()}"
+        if current_session.status == "completed":
+            status += " | ✅ Conversation completed"
+
+        return history, status
+
+    except Exception as e:
+        return history, f"❌ Error: {str(e)}"
+
+
+def export_conversation():
+    """Export the current conversation"""
+    global current_session
+
+    if not current_session:
+        return "❌ No active session to export.", None
+
+    try:
+        filepath = conversation_to_transcript(current_session)
+        return f"✅ Conversation exported to {filepath}", filepath
+    except Exception as e:
+        return f"❌ Error exporting conversation: {str(e)}", None
+
+
+def generate_conversation_summary():
+    """Generate AI summary of the current conversation"""
+    global current_session
+
+    if not current_session:
+        return "❌ No active session. Start a conversation first.", ""
+
+    if not llm_backend:
+        return "❌ LLM backend not initialized.", ""
+
+    if current_session.get_turn_count() < 3:
+        return "❌ Not enough conversation data. Have at least 2-3 exchanges first.", ""
+
+    try:
+        # Get the flow
+        flow = saved_flows.get(current_session.flow_id)
+        if not flow:
+            return "❌ Flow not found.", ""
+
+        # Create moderator and generate summary
+        moderator = ConversationModerator(llm_backend, flow)
+        summary = moderator.generate_summary(current_session)
+
+        # Format summary with stats
+        stats = current_session.get_summary_stats()
+        formatted_summary = f"""## Conversation Summary
+
+**Session Details:**
+- Session ID: {current_session.id}
+- Flow: {current_session.flow_name}
+- Total Turns: {stats['total_turns']} ({stats['user_turns']} user, {stats['ai_turns']} AI)
+- Duration: {stats['duration_minutes']} minutes
+- Status: {stats['status']}
+
+---
+
+{summary}
+
+---
+
+*Summary generated by AI. Review for accuracy.*
+"""
+
+        return "✅ Summary generated successfully!", formatted_summary
+
+    except Exception as e:
+        return f"❌ Error generating summary: {str(e)}", ""
+
+
+def update_probing_threshold(threshold: int):
+    """Update the probing threshold for follow-up questions"""
+    # This will be used when creating new moderators
+    return f"✅ Probing threshold set to every {threshold} responses"
+
+
+def get_conversation_metrics():
+    """Get real-time conversation metrics"""
+    global current_session
+
+    if not current_session:
+        return """**No Active Session**
+
+Start a conversation to see metrics."""
+
+    stats = current_session.get_summary_stats()
+    user_turns = [t for t in current_session.conversation_history if t.role == "user"]
+
+    # Calculate follow-up count (AI turns that aren't linked to nodes)
+    follow_ups = len([t for t in current_session.conversation_history
+                      if t.role == "ai" and not t.node_id])
+    scripted = stats['ai_turns'] - follow_ups
+
+    metrics_md = f"""## 📊 Live Conversation Metrics
+
+**Engagement:**
+- Total Exchanges: {stats['user_turns']}
+- User Responses: {stats['user_turns']}
+- AI Questions: {stats['ai_turns']}
+
+**Question Mix:**
+- Scripted Questions: {scripted}
+- Dynamic Follow-ups: {follow_ups}
+- Follow-up Rate: {(follow_ups / max(stats['ai_turns'], 1) * 100):.1f}%
+
+**Quality Indicators:**
+- Avg Response Length: {stats['avg_user_response_length']:.0f} characters
+- Duration: {stats['duration_minutes']} min
+- Status: {stats['status'].upper()}
+
+**Session Info:**
+- Session ID: `{current_session.id[:8]}...`
+- Flow: {current_session.flow_name}
+"""
+
+    return metrics_md
+
+
+def analyze_multiple_sessions(uploaded_files):
+    """Analyze multiple conversation sessions"""
+    if not uploaded_files:
+        return "❌ Please upload at least one conversation JSON file.", "", None
+
+    if not llm_backend:
+        return "⚠️ LLM backend not configured. Basic analysis only (no AI insights).", "", None
+
+    try:
+        # Load session data from uploaded files
+        session_data_list = []
+
+        for file in uploaded_files:
+            with open(file.name, 'r') as f:
+                data = json.load(f)
+                session_data_list.append(data)
+
+        # Create analytics instance
+        analytics = ConversationAnalytics(llm_backend)
+        loaded_count = analytics.load_sessions(session_data_list)
+
+        if loaded_count == 0:
+            return "❌ No valid sessions found in uploaded files.", "", None
+
+        # Generate comprehensive report
+        report = analytics.generate_comprehensive_report()
+
+        # Export aggregated data
+        export_data = analytics.export_aggregated_data()
+        export_file = save_json_file(export_data, "multi_session_analysis")
+
+        status = f"✅ Successfully analyzed {loaded_count} sessions from {len(uploaded_files)} files"
+
+        return status, report, export_file
+
+    except Exception as e:
+        return f"❌ Error analyzing sessions: {str(e)}", "", None
+
+
+# ===========================
+# Gradio Interface
+# ===========================
+
+def create_interface():
+    """Create the main Gradio interface"""
+
+    with gr.Blocks(
+        title="Project Echo - Qualitative Research Assistant",
+        theme=gr.themes.Soft(primary_hue="blue", secondary_hue="slate")
+    ) as app:
+
+        gr.Markdown("""
+        # Project Echo - Your AI-Powered Qualitative Research Assistant
+
+        Battle the blank page, reach global audiences, and uncover insights with AI assistance.
+        """)
+
+        # Show backend status
+        if llm_backend:
+            status_msg = f"✅ **Active LLM Provider:** {llm_backend.provider.value.upper()} | Model: {llm_backend.model}"
+            bg_color = "rgba(0, 255, 0, 0.1)"
+        else:
+            status_msg = """⚠️ **LLM Provider Not Configured**
+
+**To use this app, you need to configure an LLM provider:**
+
+1. **Easiest (HuggingFace Spaces):** Make sure your Space is PUBLIC and HF_TOKEN will be auto-available
+2. **Best Quality:** Add `OPENAI_API_KEY` in Space Settings → Variables
+3. **Alternative:** Add `ANTHROPIC_API_KEY` or `HUGGINGFACE_API_KEY`
+
+See the **About** tab for detailed instructions."""
+            bg_color = "rgba(255, 165, 0, 0.2)"
+
+        gr.Markdown(f'<div style="background-color: {bg_color}; padding: 15px; border-radius: 5px; margin: 10px 0; border-left: 4px solid #FF6B6B;">{status_msg}</div>')
+
+        with gr.Tabs() as tabs:
+
+            # ========== SURVEY GENERATION TAB ==========
+            with gr.Tab("📝 Generate Survey"):
+                gr.Markdown("""
+                ## Battle the Blank Page
+                Share an outline and get AI-powered surveys drafted in minutes,
+                complete with industry best practices.
+                """)
+
+                with gr.Row():
+                    with gr.Column(scale=1):
+                        outline_input = gr.Textbox(
+                            label="Your Survey Outline or Topic",
+                            placeholder="Example: I want to understand patient experiences with a new diabetes medication, focusing on effectiveness, side effects, and quality of life impacts.",
+                            lines=6
+                        )
+
+                        survey_type_input = gr.Radio(
+                            label="Survey Type",
+                            choices=["Qualitative", "Quantitative", "Mixed"],
+                            value="Qualitative"
+                        )
+
+                        num_questions_input = gr.Slider(
+                            label="Number of Questions",
+                            minimum=5,
+                            maximum=25,
+                            value=10,
+                            step=1
+                        )
+
+                        audience_input = gr.Textbox(
+                            label="Target Audience",
+                            placeholder="Example: Adults aged 30-65 with Type 2 diabetes",
+                            value="General audience"
+                        )
+
+                        generate_btn = gr.Button("🚀 Generate Survey", variant="primary", size="lg")
+
+                    with gr.Column(scale=1):
+                        gen_status = gr.Textbox(label="Status", interactive=False)
+                        gen_output = gr.Markdown(label="Generated Survey")
+
+                gen_download = gr.File(label="Download Survey JSON", visible=False)
+
+                # Event handlers
+                generate_btn.click(
+                    fn=generate_survey_from_outline,
+                    inputs=[outline_input, survey_type_input, num_questions_input, audience_input],
+                    outputs=[gen_status, gen_output, gen_download]
+                ).then(
+                    fn=lambda x: gr.File(value=x, visible=True) if x else gr.File(visible=False),
+                    inputs=[gen_download],
+                    outputs=[gen_download]
+                )
+
+            # ========== TRANSLATION TAB ==========
+            with gr.Tab("🌍 Translate Survey"):
+                gr.Markdown("""
+                ## Reach Global Audiences
+                Translate your surveys automatically to streamline efforts and reach wider audiences.
+                """)
+
+                with gr.Row():
+                    with gr.Column(scale=1):
+                        gr.Markdown("### Select Target Languages")
+
+                        # Create checkboxes for popular languages
+                        lang_checkboxes = gr.CheckboxGroup(
+                            label="Languages",
+                            choices=get_language_choices(),
+                            value=[]
+                        )
+
+                        translate_btn = gr.Button("🌐 Translate Survey", variant="primary", size="lg")
+
+                        gr.Markdown("""
+                        **Note:** Make sure you've generated a survey first, or upload one using the JSON format.
+                        """)
+
+                    with gr.Column(scale=1):
+                        trans_status = gr.Textbox(label="Translation Status", interactive=False)
+                        trans_output = gr.Markdown(label="Translations")
+
+                trans_download = gr.File(label="Download Translations JSON", visible=False)
+
+                # Event handlers
+                def extract_lang_codes(selected_items):
+                    """Extract language codes from checkbox selections"""
+                    return [item.split(" - ")[0] for item in selected_items]
+
+                translate_btn.click(
+                    fn=lambda x: translate_current_survey(extract_lang_codes(x)),
+                    inputs=[lang_checkboxes],
+                    outputs=[trans_status, trans_output, trans_download]
+                ).then(
+                    fn=lambda x: gr.File(value=x, visible=True) if x else gr.File(visible=False),
+                    inputs=[trans_download],
+                    outputs=[trans_download]
+                )
+
+            # ========== ANALYSIS TAB ==========
+            with gr.Tab("📊 Analyze Data"):
+                gr.Markdown("""
+                ## Uncover Key Insights
+                Upload your survey responses and get AI-assisted summaries of key findings,
+                themes, and trends.
+                """)
+
+                with gr.Row():
+                    with gr.Column(scale=1):
+                        responses_input = gr.Textbox(
+                            label="Survey Responses (JSON)",
+                            placeholder='[{"q1": "response 1", "q2": "response 2"}, ...]',
+                            lines=10
+                        )
+
+                        questions_input = gr.Textbox(
+                            label="Questions (JSON, Optional)",
+                            placeholder='[{"question_text": "What is your experience?", ...}]',
+                            lines=5
+                        )
+
+                        with gr.Row():
+                            analyze_btn = gr.Button("🔍 Analyze Data", variant="primary", size="lg")
+                            example_btn = gr.Button("Load Example", variant="secondary")
+
+                    with gr.Column(scale=1):
+                        analysis_status = gr.Textbox(label="Status", interactive=False)
+                        analysis_output = gr.Markdown(label="Analysis Report")
+
+                analysis_download = gr.File(label="Download Analysis JSON", visible=False)
+
+                # Event handlers
+                analyze_btn.click(
+                    fn=analyze_survey_data,
+                    inputs=[responses_input, questions_input],
+                    outputs=[analysis_status, analysis_output, analysis_download]
+                ).then(
+                    fn=lambda x: gr.File(value=x, visible=True) if x else gr.File(visible=False),
+                    inputs=[analysis_download],
+                    outputs=[analysis_download]
+                )
+
+                example_btn.click(
+                    fn=load_example_responses,
+                    outputs=[responses_input]
+                )
+
+            # ========== CONVERSATIONAL RESEARCH TAB ==========
+            with gr.Tab("💬 Conversational Research"):
+                gr.Markdown("""
+                ## AI-Moderated Conversations
+                Design conversation flows and conduct AI-powered qualitative interviews with respondents.
+                """)
+
+                with gr.Tabs():
+                    # Design Flow Sub-Tab
+                    with gr.Tab("🎨 Design Flow"):
+                        gr.Markdown("""
+                        ### Create Conversation Flows
+                        Design custom conversation paths for AI-moderated interviews.
+                        """)
+
+                        with gr.Row():
+                            with gr.Column(scale=1):
+                                gr.Markdown("#### Flow Setup")
+
+                                flow_name_input = gr.Textbox(
+                                    label="Flow Name",
+                                    placeholder="e.g., HCP Interview for New Dermatology Product",
+                                    value=""
+                                )
+
+                                flow_desc_input = gr.Textbox(
+                                    label="Flow Description",
+                                    placeholder="Describe the purpose of this conversation flow...",
+                                    lines=3
+                                )
+
+                                with gr.Row():
+                                    create_flow_btn = gr.Button("✨ Create New Flow", variant="primary")
+                                    load_example_flow_btn = gr.Button("📋 Load Example", variant="secondary")
+
+                                with gr.Row():
+                                    regenerate_flow_btn = gr.Button("🔄 Regenerate Flow", variant="secondary")
+                                    clear_flow_btn = gr.Button("🗑️ Clear All Steps", variant="stop")
+
+                                flow_id_state = gr.State(value="")
+
+                                gr.Markdown("#### Add Steps to Flow")
+
+                                node_content_input = gr.Textbox(
+                                    label="Question/Message",
+                                    placeholder="Enter the question or message for this step...",
+                                    lines=4
+                                )
+
+                                node_type_input = gr.Radio(
+                                    label="Step Type",
+                                    choices=["Question", "End"],
+                                    value="Question"
+                                )
+
+                                add_node_btn = gr.Button("➕ Add Step", variant="secondary")
+
+                                save_flow_btn = gr.Button("💾 Save Flow", variant="primary")
+
+                            with gr.Column(scale=1):
+                                flow_status = gr.Textbox(label="Status", interactive=False)
+                                flow_display = gr.Markdown(label="Flow Preview", value="No flow created yet")
+
+                        flow_download = gr.File(label="Download Flow JSON", visible=False)
+
+                        # Event handlers for flow design
+                        create_flow_btn.click(
+                            fn=create_new_flow,
+                            inputs=[flow_name_input, flow_desc_input],
+                            outputs=[flow_status, flow_display, flow_id_state]
+                        )
+
+                        load_example_flow_btn.click(
+                            fn=load_example_flow,
+                            outputs=[flow_status, flow_display, flow_id_state]
+                        )
+
+                        regenerate_flow_btn.click(
+                            fn=regenerate_flow_content,
+                            inputs=[flow_id_state],
+                            outputs=[flow_status, flow_display]
+                        )
+
+                        def clear_flow(flow_id):
+                            """Clear all steps from the current flow"""
+                            if not flow_id:
+                                return "❌ No flow selected.", ""
+                            flow = saved_flows.get(flow_id)
+                            if flow:
+                                flow.nodes = []
+                                return "✅ All steps cleared. You can now add new ones.", display_flow(flow)
+                            return "❌ Flow not found.", ""
+
+                        clear_flow_btn.click(
+                            fn=clear_flow,
+                            inputs=[flow_id_state],
+                            outputs=[flow_status, flow_display]
+                        )
+
+                        add_node_btn.click(
+                            fn=add_flow_node,
+                            inputs=[flow_id_state, node_content_input, node_type_input],
+                            outputs=[flow_status, flow_display]
+                        ).then(
+                            fn=lambda: "",
+                            outputs=[node_content_input]
+                        )
+
+                        save_flow_btn.click(
+                            fn=save_current_flow,
+                            inputs=[flow_id_state],
+                            outputs=[flow_status, flow_download]
+                        ).then(
+                            fn=lambda x: gr.File(value=x, visible=True) if x else gr.File(visible=False),
+                            inputs=[flow_download],
+                            outputs=[flow_download]
+                        )
+
+                    # Conduct Interview Sub-Tab
+                    with gr.Tab("🎙️ Conduct Interview"):
+                        gr.Markdown("""
+                        ### AI-Moderated Interview
+                        Start a conversation session with the AI moderator using your designed flow.
+                        """)
+
+                        with gr.Row():
+                            with gr.Column(scale=2):
+                                conversation_flow_selector = gr.State(value="")
+
+                                gr.Markdown("""
+                                **Instructions:**
+                                1. Design a flow in the 'Design Flow' tab first (or load the example)
+                                2. Configure AI moderator settings below (optional)
+                                3. Click 'Start Conversation' to begin
+                                4. The AI moderator will ask questions and adapt with follow-ups
+                                5. Generate summary and export when finished
+                                """)
+
+                                # Moderator Configuration
+                                with gr.Accordion("⚙️ AI Moderator Settings", open=False):
+                                    gr.Markdown("**Follow-up Question Configuration**")
+                                    probing_threshold_slider = gr.Slider(
+                                        label="Follow-up Frequency",
+                                        info="Ask dynamic follow-ups every N user responses",
+                                        minimum=2,
+                                        maximum=10,
+                                        value=3,
+                                        step=1
+                                    )
+                                    probing_status = gr.Textbox(label="Settings Status", interactive=False, value="Default: Every 3 responses")
+
+                                with gr.Row():
+                                    start_conversation_btn = gr.Button("🚀 Start Conversation", variant="primary", scale=2)
+                                    export_conversation_btn = gr.Button("📥 Export", variant="secondary", scale=1)
+                                    summary_btn = gr.Button("✨ Generate Summary", variant="secondary", scale=2)
+
+                                conversation_status = gr.Textbox(label="Session Status", interactive=False)
+                                conversation_download = gr.File(label="Download Transcript", visible=False)
+
+                                # Summary Display
+                                with gr.Accordion("📝 Conversation Summary", open=False):
+                                    summary_display = gr.Markdown(label="AI-Generated Summary", value="No summary yet. Complete conversation and click 'Generate Summary'.")
+
+                            with gr.Column(scale=3):
+                                chatbot = gr.Chatbot(
+                                    label="AI-Moderated Interview",
+                                    height=400
+                                )
+
+                                msg_input = gr.Textbox(
+                                    label="Your Response",
+                                    placeholder="Type your response here...",
+                                    lines=2
+                                )
+
+                                with gr.Row():
+                                    submit_btn = gr.Button("Send", variant="primary")
+                                    clear_btn = gr.Button("Clear")
+
+                                # Live Metrics Panel
+                                with gr.Accordion("📊 Live Metrics", open=True):
+                                    metrics_display = gr.Markdown(value="**No Active Session**\n\nStart a conversation to see metrics.")
+
+                        # Chat event handlers
+                        def user_submit(user_message, history):
+                            """Handle user message submission"""
+                            if not user_message:
+                                return history, history, ""
+                            return history, history + [[user_message, None]], ""
+
+                        def bot_respond(history):
+                            """Get bot response and update metrics"""
+                            if not history or history[-1][1] is not None:
+                                return history, "", get_conversation_metrics()
+
+                            user_msg = history[-1][0]
+                            updated_history, status = chat_with_moderator(user_msg, history[:-1])
+                            metrics = get_conversation_metrics()
+                            return updated_history, status, metrics
+
+                        # Probing threshold configuration
+                        probing_threshold_slider.change(
+                            fn=update_probing_threshold,
+                            inputs=[probing_threshold_slider],
+                            outputs=[probing_status]
+                        )
+
+                        # Start conversation
+                        start_conversation_btn.click(
+                            fn=lambda: saved_flows[list(saved_flows.keys())[-1]].id if saved_flows else "",
+                            outputs=[conversation_flow_selector]
+                        ).then(
+                            fn=start_conversation_session,
+                            inputs=[conversation_flow_selector],
+                            outputs=[chatbot, conversation_status]
+                        ).then(
+                            fn=get_conversation_metrics,
+                            outputs=[metrics_display]
+                        )
+
+                        # Message submission
+                        msg_input.submit(
+                            fn=user_submit,
+                            inputs=[msg_input, chatbot],
+                            outputs=[chatbot, chatbot, msg_input],
+                            queue=False
+                        ).then(
+                            fn=bot_respond,
+                            inputs=[chatbot],
+                            outputs=[chatbot, conversation_status, metrics_display]
+                        )
+
+                        submit_btn.click(
+                            fn=user_submit,
+                            inputs=[msg_input, chatbot],
+                            outputs=[chatbot, chatbot, msg_input],
+                            queue=False
+                        ).then(
+                            fn=bot_respond,
+                            inputs=[chatbot],
+                            outputs=[chatbot, conversation_status, metrics_display]
+                        )
+
+                        clear_btn.click(lambda: None, None, chatbot, queue=False)
+
+                        # Generate summary
+                        summary_btn.click(
+                            fn=generate_conversation_summary,
+                            outputs=[conversation_status, summary_display]
+                        )
+
+                        # Export conversation
+                        export_conversation_btn.click(
+                            fn=export_conversation,
+                            outputs=[conversation_status, conversation_download]
+                        ).then(
+                            fn=lambda x: gr.File(value=x, visible=True) if x else gr.File(visible=False),
+                            inputs=[conversation_download],
+                            outputs=[conversation_download]
+                        )
+
+                    # Analyze Conversations Sub-Tab
+                    with gr.Tab("📊 Analyze Conversations"):
+                        gr.Markdown("""
+                        ### Multi-Session Analysis
+                        Analyze patterns and insights across multiple conversation sessions.
+                        Upload conversation JSON files (exported from the 'Conduct Interview' tab).
+                        """)
+
+                        with gr.Row():
+                            with gr.Column(scale=1):
+                                gr.Markdown("""
+                                **How it works:**
+                                1. Conduct multiple interviews in the 'Conduct Interview' tab
+                                2. Export each conversation as JSON
+                                3. Upload all JSON files here
+                                4. Click 'Analyze Sessions' to generate comprehensive report
+                                5. Get AI-powered insights across all conversations
+
+                                **Minimum Requirements:**
+                                - At least 3-5 sessions recommended
+                                - 10+ total user responses across all sessions
+                                """)
+
+                                session_files_upload = gr.File(
+                                    label="Upload Conversation Sessions (JSON)",
+                                    file_count="multiple",
+                                    file_types=[".json"],
+                                    type="filepath"
+                                )
+
+                                analyze_sessions_btn = gr.Button("🔍 Analyze Sessions", variant="primary", size="lg")
+
+                                analytics_status = gr.Textbox(label="Analysis Status", interactive=False)
+                                analytics_download = gr.File(label="Download Analysis JSON", visible=False)
+
+                            with gr.Column(scale=1):
+                                analytics_report = gr.Markdown(
+                                    label="Multi-Session Analysis Report",
+                                    value="""# Multi-Session Analysis
+
+**Upload session files to begin analysis.**
+
+The report will include:
+- 📊 Aggregate statistics across all sessions
+- 🔑 Common keywords and topics
+- 💡 AI-powered cross-session insights
+- 📋 Individual session summaries
+- 🎯 Research recommendations
+"""
+                                )
+
+                        # Analytics event handlers
+                        analyze_sessions_btn.click(
+                            fn=analyze_multiple_sessions,
+                            inputs=[session_files_upload],
+                            outputs=[analytics_status, analytics_report, analytics_download]
+                        ).then(
+                            fn=lambda x: gr.File(value=x, visible=True) if x else gr.File(visible=False),
+                            inputs=[analytics_download],
+                            outputs=[analytics_download]
+                        )
+
+            # ========== ABOUT TAB ==========
+            with gr.Tab("ℹ️ About"):
+                gr.Markdown("""
+                ## About Project Echo
+
+                Project Echo is a comprehensive qualitative research assistant that helps you:
+
+                ### 🎯 Generate Surveys
+                - Create professional surveys from simple outlines
+                - Follow industry best practices automatically
+                - Save hours of questionnaire design time
+
+                ### 🌍 Translate Globally
+                - Reach audiences in 18+ languages
+                - Maintain cultural appropriateness
+                - Expand your research scope effortlessly
+
+                ### 📊 Analyze Results
+                - Extract key themes automatically
+                - Identify patterns and trends
+                - Generate actionable insights
+
+                ### 🔧 Configuration Guide
+
+                **For HuggingFace Spaces (Recommended):**
+
+                No configuration needed! The app automatically uses the HF Inference API with the built-in `HF_TOKEN`.
+
+                **Supported Models:**
+                - Default: `mistralai/Mixtral-8x7B-Instruct-v0.1`
+                - You can change by setting `LLM_MODEL` environment variable
+
+                **For Other LLM Providers:**
+
+                Add these environment variables in your Space Settings:
+
+                1. **OpenAI** (Best quality, paid):
+                   - `LLM_PROVIDER=openai`
+                   - `OPENAI_API_KEY=sk-your-key`
+
+                2. **Anthropic Claude** (Best reasoning, paid):
+                   - `LLM_PROVIDER=anthropic`
+                   - `ANTHROPIC_API_KEY=your-key`
+
+                3. **Custom HuggingFace Model**:
+                   - `LLM_PROVIDER=huggingface`
+                   - `LLM_MODEL=your-model-name`
+
+                **💡 Pro Tip:** For production use, we recommend OpenAI or Anthropic for faster, more reliable results.
+
+                **Supported LLM Providers:**
+                - HuggingFace Inference API (Free tier available)
+                - OpenAI (GPT-4, GPT-4o-mini, GPT-3.5)
+                - Anthropic (Claude 3.5 Sonnet, Claude 3 Opus)
+                - LM Studio (local development only)
+
+                ### 📄 Data Privacy
+
+                - All processing is done through your configured LLM provider
+                - No data is stored permanently by this application
+                - Survey data and responses remain in your control
+
+                ### 🚀 Getting Started
+
+                1. **Generate** a survey from your research outline
+                2. **Translate** it to reach global audiences
+                3. Collect responses from participants
+                4. **Analyze** the data to uncover insights
+
+                ---
+
+                Built with ❤️ using Gradio and state-of-the-art LLMs
+                """)
+
+        return app
+
+
+# ===========================
+# Main Entry Point
+# ===========================
+
+if __name__ == "__main__":
+    demo = create_interface()
+
+    # Launch with appropriate settings
+    demo.launch(
+        server_name="0.0.0.0",  # Allow external access
+        server_port=7860,  # Standard HF Spaces port
+        share=False,  # Don't create a public link (HF Spaces handles this)
+        show_error=True
+    )

conversation_flow.py

@@ -141,6 +141,96 @@ class ConversationFlow:
 
         return True, "Flow is valid"
 
+    def generate_flow_with_ai(self, llm_backend, num_questions: int = 5):
+        """
+        Generate conversation flow nodes using AI based on flow name and description.
+
+        Args:
+            llm_backend: LLM backend to use for generation
+            num_questions: Number of conversation steps to generate
+        """
+        if not self.name or not self.description:
+            raise ValueError("Flow must have a name and description to generate nodes")
+
+        # Build prompt for generating conversation flow
+        prompt = f"""Task: Design a structured conversation flow
+
+**Interview Topic:** {self.name}
+
+**Interview Purpose:** {self.description}
+
+**Your Task:** Create {num_questions} conversation steps for a structured qualitative research interview.
+
+**Guidelines for Each Step:**
+- Start with an opening that builds rapport and explains the purpose
+- Progress from general to specific questions
+- Each step should be clear, open-ended, and encourage detailed responses
+- Include natural transition phrases
+- End with a closing that thanks the respondent
+- Make questions natural and conversational, not robotic
+
+**Output Format:** Number each step (1., 2., 3., etc.) with the exact question or statement to use.
+
+**Generate {num_questions} Interview Steps:**
+
+1."""
+
+        messages = [
+            {
+                "role": "system",
+                "content": "You are an expert qualitative research interviewer designing a conversation flow. Create engaging, professional interview questions that will elicit detailed, meaningful responses."
+            },
+            {"role": "user", "content": prompt}
+        ]
+
+        try:
+            response = llm_backend.generate(messages, max_tokens=1500, temperature=0.7)
+            self._parse_and_add_nodes(response)
+            return True, "Flow generated successfully!"
+        except Exception as e:
+            return False, f"Flow generation failed: {str(e)}"
+
+    def _parse_and_add_nodes(self, response: str):
+        """
+        Parse LLM response and create conversation nodes.
+
+        Args:
+            response: The LLM-generated response containing numbered questions
+        """
+        import re
+
+        # Pattern to match numbered items: "1. Question" or "1) Question"
+        pattern = r'\d+[\.\)]\s+(.+?)(?=\d+[\.\)]|\Z)'
+        matches = re.findall(pattern, response, re.DOTALL)
+
+        if not matches:
+            # Fallback: split by lines and look for question-like content
+            lines = response.split('\n')
+            matches = [line.strip() for line in lines if line.strip() and len(line.strip()) > 20]
+
+        for i, match in enumerate(matches):
+            # Clean up the match
+            content = match.split('\n')[0].strip()
+
+            if not content or len(content) < 10:
+                continue
+
+            # Determine node type
+            node_type = "question"
+            if i == 0:
+                node_type = "opening"
+            elif i == len(matches) - 1:
+                node_type = "end"
+
+            # Create and add node
+            node = ConversationNode(content=content, node_type=node_type)
+
+            if self.nodes:
+                # Link to previous node
+                self.nodes[-1].next = node.id
+
+            self.add_node(node)
+
 
 def create_example_flow() -> ConversationFlow:
     """Create an example conversation flow"""

conversation_moderator.py

@@ -105,21 +105,33 @@ class ConversationModerator:
         Returns:
             A follow-up question
         """
-        # Create prompt for generating follow-up
-        system_prompt = """You are a professional qualitative research interviewer. Your goal is to probe deeper into the respondent's answers to uncover insights.
+        # Create prompt for generating follow-up - optimized for Mistral/Mixtral
+        system_prompt = """You are a skilled qualitative research interviewer conducting a professional interview. Your role is to:
+- Build trust and rapport with respondents
+- Probe deeper into meaningful points they raise
+- Encourage detailed, thoughtful responses
+- Stay curious and engaged without bias
 
-Generate ONE follow-up question that:
-- Explores an interesting point the respondent mentioned
-- Asks for more detail or clarification
-- Uses phrases like "Tell me more about...", "Can you elaborate on...", "What do you mean by...", "Why do you think..."
-- Is empathetic and non-judgmental
-- Is concise (one sentence)
+When generating follow-up questions:
+- Focus on a single interesting or important point they mentioned
+- Ask for more detail, clarity, or deeper thinking
+- Use natural, conversational phrasing
+- Show genuine interest in their perspective
+- Keep questions clear and concise (one sentence)
+- Be empathetic and non-judgmental
 
-Respond ONLY with the follow-up question, nothing else."""
+Output ONLY the follow-up question text, with no additional explanation or commentary."""
 
-        user_prompt = f"""The respondent just said: "{user_message}"
+        user_prompt = f"""**Respondent's Statement:** "{user_message}"
 
-Generate a single follow-up question to probe deeper into their response."""
+**Task:** Generate one thoughtful follow-up question that probes deeper into what they said.
+
+Focus on:
+- Exploring an interesting or important point
+- Asking for more detail or their reasoning
+- Encouraging reflection and deeper thinking
+
+Provide ONLY the follow-up question text."""
 
         messages = [
             {"role": "system", "content": system_prompt},
@@ -185,19 +197,29 @@ Generate a single follow-up question to probe deeper into their response."""
 
         transcript = "\n".join(transcript_parts)
 
-        system_prompt = """You are analyzing a qualitative research interview. Generate a concise summary that captures:
-1. The main topics discussed
-2. Key insights or themes from the respondent
-3. Notable quotes or moments
-4. Overall sentiment
+        system_prompt = """You are a qualitative research analyst summarizing a conducted interview. Your summary should be:
+- Professional and objective
+- Grounded in what the respondent actually said
+- Organized by themes and key points
+- Include representative quotes
+- Highlight insights and implications
+- Suitable for a research report or case study"""
 
-Keep the summary to 3-4 paragraphs."""
+        user_prompt = f"""Task: Summarize this qualitative research interview
 
-        user_prompt = f"""Summarize this interview:
+**Interview Transcript:**
 
 {transcript}
 
-Provide a professional summary suitable for a research report."""
+**Summary Requirements:**
+1. **Main Topics:** What topics or subjects did the respondent discuss?
+2. **Key Insights:** What are the most important or revealing points they made?
+3. **Themes:** What patterns or recurring themes emerge from their responses?
+4. **Representative Quotes:** Include 2-3 direct quotes that capture important moments
+5. **Sentiment & Tone:** What is the overall emotional tone and sentiment?
+
+**Format:** Write a professional summary of 3-4 paragraphs suitable for a research report.
+Start with a brief overview, then discuss key themes and insights."""
 
         messages = [
             {"role": "system", "content": system_prompt},

data_analyzer.py

@@ -73,23 +73,26 @@ class DataAnalyzer:
         for i, resp in enumerate(sample_responses, 1):
             context += f"{i}. {resp[:200]}...\n"  # Truncate long responses
 
-        prompt = f"""Analyze the following survey responses and provide an executive summary.
+        prompt = f"""Task: Analyze survey responses and generate an executive summary
 
 {context}
 
-Provide a summary that includes:
-1. Overview: High-level summary of what the data shows (2-3 sentences)
-2. Key patterns: Main patterns or trends observed
-3. Notable findings: Interesting or unexpected discoveries
-4. Response quality: Assessment of response depth and engagement
+**Your Analysis Should Include:**
 
-Respond with a JSON object with these fields:
+1. **Overview:** A clear, concise high-level summary of what the data reveals (2-3 sentences)
+2. **Key Patterns:** Main patterns, trends, or recurring themes observed across responses
+3. **Notable Findings:** Interesting, surprising, or unexpected discoveries in the data
+4. **Response Quality:** Assessment of how thoughtful, engaged, and detailed the responses are
+
+**Output Format:** Respond ONLY with valid JSON:
 {{
-  "overview": "...",
-  "key_patterns": ["pattern 1", "pattern 2", ...],
-  "notable_findings": ["finding 1", "finding 2", ...],
-  "response_quality": "..."
-}}"""
+  "overview": "Clear summary of overall findings",
+  "key_patterns": ["pattern 1", "pattern 2", "pattern 3"],
+  "notable_findings": ["surprising finding 1", "unexpected discovery"],
+  "response_quality": "Assessment of engagement level"
+}}
+
+**Important:** Ensure your response is valid JSON that can be parsed. Do not include any text outside the JSON object."""
 
         messages = [
             {"role": "system", "content": self._get_analyst_system_prompt()},
@@ -113,26 +116,32 @@ Respond with a JSON object with these fields:
         sample_size = min(100, len(response_texts))
         sample_responses = response_texts[:sample_size]
 
-        prompt = f"""Analyze the following {len(sample_responses)} survey responses and identify the top {num_themes} themes.
+        prompt = f"""Task: Extract and analyze themes from survey responses
+
+**Data:** Analyzing {len(sample_responses)} survey responses
 
 Responses:
 {self._format_responses_for_prompt(sample_responses)}
 
-For each theme, provide:
-1. Theme name: A short, descriptive name
-2. Description: What this theme represents
-3. Prevalence: Estimated percentage of responses mentioning this theme
-4. Example quotes: 2-3 representative quotes from the responses
+**Your Task:** Identify the top {num_themes} distinct themes that emerge from these responses.
 
-Respond with a JSON array of theme objects:
+**For Each Theme, Provide:**
+1. **Theme Name:** A short, memorable, and descriptive label
+2. **Description:** Clear explanation of what this theme represents and its significance
+3. **Prevalence:** Estimated percentage of responses that mention or relate to this theme
+4. **Example Quotes:** 2-3 actual, representative quotes from responses that illustrate this theme
+
+**Output Format:** Respond ONLY with a valid JSON array:
 [
   {{
-    "theme_name": "...",
-    "description": "...",
+    "theme_name": "Clear, concise theme label",
+    "description": "What this theme means and why it matters",
     "prevalence": "XX%",
-    "example_quotes": ["quote 1", "quote 2"]
+    "example_quotes": ["exact quote from responses", "another quote"]
   }}
-]"""
+]
+
+**Important:** Ensure all responses are valid JSON. Do not include text outside the array."""
 
         messages = [
             {"role": "system", "content": self._get_analyst_system_prompt()},
@@ -159,28 +168,34 @@ Respond with a JSON array of theme objects:
         sample_size = min(100, len(response_texts))
         sample_responses = response_texts[:sample_size]
 
-        prompt = f"""Analyze the sentiment of these {len(sample_responses)} survey responses.
+        prompt = f"""Task: Analyze sentiment across survey responses
+
+**Data:** Analyzing sentiment in {len(sample_responses)} survey responses
 
 Responses:
 {self._format_responses_for_prompt(sample_responses)}
 
-Provide sentiment analysis including:
-1. Overall sentiment: positive, negative, neutral, or mixed
-2. Sentiment distribution: Estimated percentage breakdown
-3. Emotional tone: Key emotions detected
-4. Intensity: How strong the sentiments are
+**Your Task:** Conduct a comprehensive sentiment analysis of these responses.
+
+**Analysis Should Include:**
+1. **Overall Sentiment:** The dominant sentiment tone (positive, negative, neutral, or mixed)
+2. **Sentiment Distribution:** Estimated percentage breakdown across sentiment categories
+3. **Emotional Tone:** Key emotions or emotional themes detected in responses
+4. **Intensity:** The strength of the sentiments (low, moderate, or high)
 
-Respond with JSON:
+**Output Format:** Respond ONLY with valid JSON:
 {{
-  "overall_sentiment": "...",
+  "overall_sentiment": "positive|negative|neutral|mixed",
   "distribution": {{
     "positive": "XX%",
     "neutral": "XX%",
     "negative": "XX%"
   }},
-  "emotions": ["emotion1", "emotion2", ...],
+  "emotions": ["emotion1", "emotion2", "emotion3"],
   "intensity": "low|moderate|high"
-}}"""
+}}
+
+**Important:** Return only valid JSON. Do not include explanatory text."""
 
         messages = [
             {"role": "system", "content": self._get_analyst_system_prompt()},
@@ -215,19 +230,25 @@ Respond with JSON:
 Sample responses:
 {self._format_responses_for_prompt(sample_responses)}
 
-Based on this data, provide 5-7 key insights that would be valuable for:
-- Understanding the target audience
-- Identifying opportunities or challenges
-- Informing strategic decisions
-- Recognizing patterns or trends
+**Task:** Extract key insights from this survey data
+
+**Generate 5-7 actionable insights** that address:
+- Understanding the target audience and their needs
+- Identifying opportunities for growth or improvement
+- Recognizing challenges or pain points
+- Understanding patterns, trends, and correlations
+- Informing strategic or product decisions
+
+**Insight Quality Criteria:**
+- **Specific:** Clear, concrete statements based on actual data patterns
+- **Actionable:** Can be used to inform decisions or actions
+- **Evidence-based:** Grounded in what respondents actually said
+- **Concise:** Clear and to the point (1-2 sentences each)
 
-Each insight should be:
-- Specific and actionable
-- Supported by the data
-- Clear and concise
+**Output Format:** Respond ONLY with a valid JSON array of insight strings:
+["Clear, actionable insight from the data", "Another specific insight", ...]
 
-Respond with a JSON array of insight strings:
-["insight 1", "insight 2", ...]"""
+**Important:** Return only JSON array. Do not include explanatory text."""
 
         messages = [
             {"role": "system", "content": self._get_analyst_system_prompt()},

llm_backend.py

@@ -1,5 +1,5 @@
 """
-LLM Backend for ConversAI - Supports multiple providers
+LLM Backend for Project Echo - Supports multiple providers
 """
 import os
 import requests
@@ -74,19 +74,28 @@ class LLMBackend:
         self.device = None
 
     def _get_default_model(self) -> str:
-        """Get default model for each provider"""
+        """Get default model for each provider with fallback chain"""
         defaults = {
             LLMProvider.OPENAI: "gpt-4o-mini",
             LLMProvider.ANTHROPIC: "claude-3-5-sonnet-20241022",
-            # Using Microsoft Phi-2 - causal LM, better at creative text generation
-            # Phi-2 is 2.7GB and instruction-tuned for following prompts
-            # Alternative: google/gemma-2b-it (2GB) or TinyLlama/TinyLlama-1.1B-Chat-v1.0 (1.1GB)
-            # NOTE: Flan-T5 models don't work well - they copy examples instead of generating
-            LLMProvider.HUGGINGFACE: "microsoft/phi-2",
+            # Preferred: Mistral-7B (better instruction following, higher quality)
+            # Fallback chain for HF Inference API if primary is gated/unavailable
+            LLMProvider.HUGGINGFACE: "mistralai/Mistral-7B-Instruct-v0.1",
             LLMProvider.LM_STUDIO: "google/gemma-3-27b"
         }
         return os.getenv("LLM_MODEL", defaults[self.provider])
 
+    def get_fallback_models(self) -> List[str]:
+        """Get fallback model chain for HF Inference API"""
+        if self.provider == LLMProvider.HUGGINGFACE:
+            return [
+                "mistralai/Mistral-7B-Instruct-v0.1",  # Primary
+                "mistralai/Mixtral-8x7B-Instruct-v0.1",  # Fallback 1: Better quality
+                "google/gemma-7b-it",  # Fallback 2: Smaller, faster
+                "microsoft/phi-2",  # Fallback 3: Original
+            ]
+        return [self.model]
+
     def _get_api_url(self) -> str:
         """Get API URL for each provider"""
         if self.provider == LLMProvider.OPENAI:
@@ -222,39 +231,63 @@ class LLMBackend:
         print(f"Model loaded successfully!")
 
     def _generate_huggingface(self, messages, max_tokens, temperature) -> str:
-        """Generate using local transformers model"""
-        # Load model if not already loaded
-        self._load_local_model()
-
-        # Convert messages to prompt
-        prompt = self._messages_to_prompt(messages)
-
-        # Tokenize input
-        inputs = self.tokenizer(prompt, return_tensors="pt", truncation=True, max_length=512)
-        inputs = inputs.to(self.device)
-
-        # Generate
-        with torch.no_grad():
-            outputs = self.local_model.generate(
-                **inputs,
-                max_new_tokens=max_tokens,
-                temperature=temperature,
-                do_sample=temperature > 0,
-                top_p=0.9,
-                pad_token_id=self.tokenizer.eos_token_id
-            )
-
-        # Decode output
-        generated_text = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
-
-        # For T5 models, the output is just the generated text
-        # For causal models, we need to remove the input prompt
-        if "t5" not in self.model.lower() and "flan" not in self.model.lower():
-            # Remove the input prompt from output
-            if generated_text.startswith(prompt):
-                generated_text = generated_text[len(prompt):].strip()
-
-        return generated_text
+        """Generate using local transformers model with fallback chain"""
+        # Try to load and generate with fallback chain
+        fallback_models = self.get_fallback_models()
+        last_error = None
+
+        for model_to_try in fallback_models:
+            try:
+                # Temporarily set model for this attempt
+                original_model = self.model
+                self.model = model_to_try
+                self.tokenizer = None  # Reset tokenizer cache
+                self.local_model = None  # Reset model cache
+
+                # Load model if not already loaded
+                self._load_local_model()
+
+                # Convert messages to prompt
+                prompt = self._messages_to_prompt(messages)
+
+                # Tokenize input
+                inputs = self.tokenizer(prompt, return_tensors="pt", truncation=True, max_length=512)
+                inputs = inputs.to(self.device)
+
+                # Generate
+                with torch.no_grad():
+                    outputs = self.local_model.generate(
+                        **inputs,
+                        max_new_tokens=max_tokens,
+                        temperature=temperature,
+                        do_sample=temperature > 0,
+                        top_p=0.9,
+                        pad_token_id=self.tokenizer.eos_token_id
+                    )
+
+                # Decode output
+                generated_text = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
+
+                # For T5 models, the output is just the generated text
+                # For causal models, we need to remove the input prompt
+                if "t5" not in self.model.lower() and "flan" not in self.model.lower():
+                    # Remove the input prompt from output
+                    if generated_text.startswith(prompt):
+                        generated_text = generated_text[len(prompt):].strip()
+
+                # Success! Update the default model for future use
+                self.model = model_to_try
+                print(f"✓ Successfully using model: {model_to_try}")
+                return generated_text
+
+            except Exception as e:
+                last_error = e
+                print(f"⚠ Model {model_to_try} failed: {str(e)[:100]}")
+                self.model = original_model  # Restore original
+                continue
+
+        # All fallbacks failed
+        raise Exception(f"All HuggingFace models failed. Last error: {str(last_error)}")
 
     def _generate_lm_studio(self, messages, max_tokens, temperature) -> str:
         """Generate using LM Studio local API"""

survey_generator.py

@@ -78,29 +78,30 @@ class SurveyGenerator:
             return f"{topic} Survey"
 
     def _get_system_prompt(self) -> str:
-        """System prompt for survey generation"""
-        return """You are a professional survey designer. Generate clear, professional survey questions."""
+        """System prompt for survey generation - optimized for Mistral/Mixtral"""
+        return """You are an expert survey designer specializing in qualitative research. Your role is to create clear, professionally-written, and contextually relevant survey questions that elicit detailed responses from respondents."""
 
     def _build_generation_prompt(self, outline, survey_type, num_questions, target_audience) -> str:
-        """Build the user prompt for survey generation"""
-        # For causal LMs (Phi, Gemma, etc.) - more direct and explicit
-        return f"""Task: Create a {survey_type} research survey
+        """Build the user prompt for survey generation - optimized for Mistral/Mixtral"""
+        return f"""You are creating a {survey_type.lower()} research survey.
 
-Research Topic: {outline}
+**Research Focus:** {outline}
 
-Target Audience: {target_audience}
+**Target Participants:** {target_audience}
 
-Create exactly {num_questions} survey questions.
+**Your Task:** Generate exactly {num_questions} high-quality survey questions.
 
-Requirements:
-- Each question must be clear, specific, and relevant to the topic
-- Questions should be appropriate for the target audience
-- Avoid yes/no questions in qualitative surveys
-- Make questions open-ended to encourage detailed responses
+**Quality Requirements:**
+- Each question must be directly relevant to the research focus
+- Questions should be specific enough to guide responses but open enough to capture diverse perspectives
+- For {survey_type.lower()} surveys: Use open-ended questions that encourage detailed, thoughtful responses
+- Avoid leading questions, double questions, or jargon that may confuse respondents
+- Ensure questions are appropriate for the target audience's knowledge and context
+- Progress from general to specific topics when possible
 
-Format: Use numbered list (1., 2., 3., etc.)
+**Format:** Output as a numbered list (1. Question text 2. Question text, etc.)
 
-Here are the {num_questions} survey questions:
+**Output {num_questions} Survey Questions:**
 
 1."""
 
@@ -277,7 +278,7 @@ Here are the {num_questions} survey questions:
 
     def refine_question(self, question: str, improvement_type: str = "clarity") -> str:
         """
-        Refine a single survey question.
+        Refine a single survey question - optimized for Mistral/Mixtral
 
         Args:
             question: The question to improve
@@ -286,19 +287,32 @@ Here are the {num_questions} survey questions:
         Returns:
             Improved question text
         """
-        prompt = f"""Improve the following survey question for better {improvement_type}:
+        improvement_guidance = {
+            "clarity": "Makes the question clearer and easier for respondents to understand without ambiguity",
+            "neutrality": "Removes any bias, leading language, or assumptions that could influence responses",
+            "specificity": "Makes the question more specific and actionable while remaining open-ended"
+        }
+
+        guidance = improvement_guidance.get(improvement_type, improvement_guidance["clarity"])
+
+        prompt = f"""Task: Improve a survey question
+
+**Original Question:** "{question}"
 
-Original Question: {question}
+**Improvement Type:** {improvement_type.title()}
 
-Provide an improved version that:
-- {"Is clearer and easier to understand" if improvement_type == "clarity" else ""}
-- {"Removes bias and leading language" if improvement_type == "neutrality" else ""}
-- {"Is more specific and actionable" if improvement_type == "specificity" else ""}
+**Your Goal:** Rewrite this question so that it {guidance}.
 
-Respond with only the improved question text, no explanation."""
+**Guidelines:**
+- Keep the question focused on a single topic
+- Use simple, clear language appropriate for the target audience
+- Avoid assumptions or leading language
+- Ensure the question can elicit meaningful responses
+
+Provide ONLY the improved question text. Do not include explanations or alternative versions."""
 
         messages = [
-            {"role": "system", "content": "You are an expert survey question designer."},
+            {"role": "system", "content": "You are an expert survey question designer with deep experience in qualitative research methodology."},
             {"role": "user", "content": prompt}
         ]
 
@@ -306,7 +320,7 @@ Respond with only the improved question text, no explanation."""
 
     def add_follow_up_questions(self, base_question: str, num_follow_ups: int = 3) -> List[str]:
         """
-        Generate follow-up questions for deeper exploration.
+        Generate follow-up questions for deeper exploration - optimized for Mistral/Mixtral
 
         Args:
             base_question: The main question
@@ -315,35 +329,48 @@ Respond with only the improved question text, no explanation."""
         Returns:
             List of follow-up question texts
         """
-        prompt = f"""Generate {num_follow_ups} follow-up questions for this main question:
+        prompt = f"""Task: Generate probing follow-up questions
+
+**Main Question:** {base_question}
 
-Main Question: {base_question}
+**Your Task:** Create {num_follow_ups} thoughtful follow-up questions that probe deeper into the respondent's answer.
 
-The follow-up questions should:
-1. Probe deeper into the topic
-2. Explore different aspects or dimensions
-3. Encourage detailed responses
-4. Follow a logical progression
+**Quality Criteria for Follow-ups:**
+1. Each question should explore a different aspect, dimension, or implication of the main topic
+2. Questions should encourage more detailed, nuanced responses
+3. Follow a logical progression from the main question
+4. Build on what a respondent might answer to the main question
+5. Each should be specific but open-ended
 
-Respond with a JSON array of question strings."""
+**Format:** Number each question (1., 2., 3., etc.)
+
+**Output {num_follow_ups} Follow-up Questions:**
+
+1."""
 
         messages = [
-            {"role": "system", "content": "You are an expert in qualitative research interviews."},
+            {"role": "system", "content": "You are an expert qualitative research interviewer skilled at designing probing questions that uncover deeper insights and nuances."},
             {"role": "user", "content": prompt}
         ]
 
         response = self.llm.generate(messages, max_tokens=500, temperature=0.7)
 
-        try:
-            # Extract JSON array
-            if "[" in response:
-                start = response.find("[")
-                end = response.rfind("]") + 1
-                follow_ups = json.loads(response[start:end])
-                return follow_ups[:num_follow_ups]
-        except:
-            pass
-
-        # Fallback: split by newlines
+        # Parse the response for follow-up questions
+        import re
+
+        # Try numbered list format first
+        pattern = r'\d+[\.\)]\s+(.+?)(?=\d+[\.\)]|\Z)'
+        matches = re.findall(pattern, response, re.DOTALL)
+
+        if matches:
+            follow_ups = [m.split('\n')[0].strip() for m in matches if m.strip()][:num_follow_ups]
+            # Ensure all end with question mark
+            follow_ups = [q if q.endswith('?') else q + '?' for q in follow_ups]
+            if follow_ups:
+                return follow_ups
+
+        # Fallback: split by newlines and look for questions
         lines = [line.strip() for line in response.split("\n") if line.strip()]
-        return [line.lstrip("0123456789.-) ") for line in lines if "?" in line][:num_follow_ups]
+        follow_ups = [line.lstrip("0123456789.-) ") for line in lines if "?" in line][:num_follow_ups]
+
+        return follow_ups if follow_ups else [f"Can you elaborate on {base_question.lower()}?" for _ in range(num_follow_ups)]